CHAPTER-4 FRUIT QUALITY GRADATION USING SHAPE, SIZE AND DEFECT ATTRIBUTES

Transcription

1 CHAPTER-4 FRUIT QUALITY GRADATION USING SHAPE, SIZE AND DEFECT ATTRIBUTES In addition to colour based estimation of apple quality, various models have been suggested to estimate external attribute based apple quality including fruit shape, size and bruise size features under the realm of machine vision based inspection of apple crop of red delicious variety. However, external attribute based available methods suffer from the limitation of being less accurate and/or inconsistent with regard to variation in fruit surface illumination, which is of course unavoidable in real time situation. An extensive experimental study is carried out in the present work to prove this fact. Results of this study are quite encouraging. From the results of experimental investigation, it is established that fruit surface illumination is one of the important parameters, for estimation of apple quality using external attributes including shape, size and bruise size features. 4.1 Introduction Fruit Shape Based Gradation Shape is considered to be one of the important external attributes in grading of apple crop as consumer choice is always to have a fruit with perfect shape (Costa et al, 2011). Even sometimes fruit with uneven shape is mistaken as defective by consumers as they think it is deformed during transportation due to improper packing. Thus in packaging industry, grading of fruits based on shape is one of the major task that is performed. A new technique is proposed to evaluate the fruit shape quantitatively using 94

2 attractor, fractal dimension and neural networks (Morimoto et al, 2000). The features which can effectively define the shape of apples are as under: (a) Heywood Circularity Factor (HCF) Circularity of a particle is a measure of irregularity or the difference from the perfect circle. Very irregular shape has circularity factor near to 0. And this factor is very useful in the area where perfectly circular objects or particles are required for their work. The area of a particle cannot indicate whether particles are lines or not because the lines are not of a known size. Straight lines such as those in Figure 4.1 (a) have small widths and low moments about their axes of orientation. However, curved lines such those in Figure 4.1 (b) have neither of these properties. Thus particle size measurements cannot be used to recognize lines. Another well-known particle measurement used for shape analysis is the Heywood circularity factor (NI Vision Concepts, 2008), which tells about how circular a particle is, and can be defined as the ratio of the particle perimeter to the perimeter of a circle of the same area. A perfect circle would have a Heywood circularity factor of 1. Longer particles would tend to have a high Heywood circularity factor. The Heywood circularity factor can be used to classify particles such as those in Figure 4.1 (c) that are not very distorted. The particle shown in Figure 4.1 (d) has a very high Heywood circularity factor but yet it is more a circle than a line and illustrates a case where the circularity factor fails to classify a shape because of the large perimeter to area ratio. Heywood circularity factor (HCF) is mathematically expressed as HCF = P/ (2 A) (4.1) where, P is particle perimeter and A is particle area 95

3 Figure 4.1 Different shapes for particle analysis (NI Vision concepts, 2008) (b) Equivalent Ellipse Axis Ratio (EEAR) EEAR is the ratio of the major axis of the equivalent ellipse to its minor axis, which is defined as the more elongated the equivalent ellipse, the higher the ellipse ratio the closer the equivalent ellipse is to a circle, the closer the ellipse ratio, is to 1 (NI Vision Concepts, 2008). = (4.2) where, a is the length of the bigger axis of the ellipse which has the same area as that of the particle defined by equation: = + + (4.3) 96

4 where, P is the particle perimeter and A, is the particle area, and b is the length of the smaller axis of the ellipse which has the same area as that of the particle defined by the equation: = + (4.4) Fruit Size Based Gradation Size is another external attribute as consumer choice is always to have fruits with equal size. In fact, this is the basis of sorting of fruits based on their size. Proper sorting of fruit ensures uniformity in fruit size, reduce packaging and transportation costs and also provides an optimum packaging configuration. Thus in packaging industry, grading of fruits based on size is one of the important task that is performed. Three methods are reported for apple size determination by applying known geometrical models (Sudhakara et al, 2002). The features which can effectively define the size of apples are as under: (a) Perimeter Various methods like area measurement, linear regression relations between the areas taken from two different viewpoints, measurement comparison of width, length, perimeter etc., are being chosen to estimate and compare the fruit size. The perimeter may be defined as the arc length of a spatially sampled curve (NI Vision Concepts, 2008). Initially, the edge of the object is identified using free noise binary image. The perimeter values of fruit are estimated with respect to the pixels values. By counting the number of pixels of the boundary that have been identified, the perimeter of the fruit can be estimated. 97

5 (b) Hydraulic Radius Hydraulic Radius is a term which plays an important role to analyze the quality of the fruit with its area and perimeter (NI Vision Concepts, 2008). To calculate the Hydraulic radius of an object in the image firstly extract the colour component and then converts into binary image using threshold technique and apply the particle analysis to calculate the radius A disk with radius R has a hydraulic radius equal to = = (4.5) Fruit Defect Based Gradation Apple surface defects is another important external attribute that is of great concern to apple producers as consumers assess apple quality by the appearance of fruit and always free from defects on surface. Apple producers always wish to get the best of their produce. In order to do so utmost care is required on their part so that their produce is free from any type of defects. Defects are considered as grade lowering characteristics. Though a number of defects can occur in the fruit but bruise is the defect that mainly occurs due to poor and negligible post-harvest handling. Reviews of the various techniques used in automatic inspection of fruits using machine vision techniques for defects identification in different fruits and also about the removal of noises is reported. (Devi et al, 2014). An automated apple surface defect sorting experimental system based on computer image technology is reported (Li et al, 2002). The methods, including image background removal, defects segmentation and identification for stem-end and calyx areas, are developed. In another work a grading method applied to Jonagold apples is reported (Leemans et al, 2004). The images are 98

6 segmented and the features of the defects are extracted. A fruit surface defect detection and classification algorithm based on attention model is reported (Panli, 2012). The same features as used for size grading can be used for defect grading of apples 4.2 Materials and Methods Three human experts are given two hundred samples and 25 samples in each grade are selected based on shape of apples. Twenty samples in each grade are taken as reference samples that are commonly agreed upon by all the experts. The images are again taken using same experimental setup illustrated in Figure 4.1 following similar procedure as adopted in chapter 2 for fruit colour samples. Data is then processed and analyzed using particle classification interface available in NI Vision Builder for Automated Inspection (NI Vision Builder, 2009; Jalili et al, 2013). Next, the human experts are again given two hundred samples and 25 samples in each grade are selected based on size of the apples. Twenty samples in each grade are taken as reference samples that are commonly agreed upon by all the experts. The images are again taken using same experimental setup as done for colour samples as shown in Figure 4.2. Data is then processed and analyzed using particle classification interface available in NI Vision Builder for Automated Inspection. Finally the whole process of sample selection by human experts, image taking, processing and analyzing is repeated for Bruise size (Defect) using experimental setup shown in Figure 4.2. The Image Data Base prepared in this chapter is further used in subsequent chapters also. 99

7 HCF Based Fruit Shape Features Fruit Shape Based Gradation EEAR Based Fruit Shape Features Fruit Shape Models Fruit Shape Classifier Image Acquisition Preprocessing including image segmentation Perimeter Based Fruit Size Features Hydraulic Radius Based Fruit Size Features Fruit Size Models Fruit Size Based Gradation Fruit Size Classifier Features Extraction Perimeter Based Bruise Size Features Hydraulic Radius Based Bruise Size Features Bruise Size Models Fruit Shape Based Gradation Fruit Defect Classifier Figure 4.2 External Fruit Attribute Based Gradation 4.3 Results and Discussion An experiment is carried out to study the consistency of fruit shape, fruit size and bruise size models with regard to estimation of apple quality under the effect of varying fruit surface illumination Evaluation of Fruit Shape Models (a) Heywood Circulatory Factor (HCF) Based Fruit Shape Model As indicated in Table-4.1, with decrease in intensity from 486 Lux to 170 Lux, the value of Heywood Circulatory Factor (HCF) decreases for all the grades. It has been established that for grades A and B, HCF is not at 310 Lux and 486 Lux. At the same time, it is conflicting at intensities 170 Lux, 253 Lux and 405 Lux. 100

8 Maximum is at 405 Lux. No with Grade B at intensities 310 Lux and 486 Lux. However a gap is there between values for Heywood Circulatory Factor between grade A & B at intensities 310 Lux and 486 Lux. Maximum gap in value of Heywood Circulatory Factor from grade A to Grade B is for intensity at 486 Lux and minimum is for intensity at 310 Lux. Heywood Circulatory Factor for grade B & C are conflicting at all intensities except 405 Lux. However, maximum is at 486 Lux and minimum is at 310 Lux. However a gap is there between values for Heywood Circulatory Factor between grade B & C at intensity 405 Lux which is very high. Heywood Circulatory Factor for grade C & D is conflicting at all intensities including 310 Lux. Maximum conflict is at 405 Lux and minimum is at 310 Lux. These results are very clearly shown as graphs in Figure 4.3 for Heywood Circulatory Factor. Also shown in Table-4.2 is the distinction possibility between consecutive grades on the basis of Heywood Circulatory Factor. Table-4.1 Range of Heywood Circulatory Factor of Red Delicious Apple Intensity Heywood Circulatory Factor (Lux) Grade-A Grade-B Grade-C Grade-D Table-4.2 Distinction Possibility Between Consecutive Grades on the Basis of Heywood Circulatory Factor (HCF) Intensity (Lux) Grades A & B Grades B & C Grades C & D 486 Yes No No 405 No Yes No 310 Yes No No 253 No No No 170 No No No 101

9 Figure 4.3 Overlapping of Heywood Circulatory Factor Range of Red Delicious Apple for Consecutive Grades at Different Intensities (Indicative graph, however, not to the exact scale along x-axis) (b) Equivalent Ellipse Axes Ratio (EEAR) Based Fruit Shape Model As indicated in Table-4.3, with decrease in intensity from 486 Lux to 170 Lux, the value of Equivalent Ellipse Axes Ratio (EEAR) decreases for all the grades. It has been established that for grades A and B, EEAR is not at 170 Lux, 310 Lux and 486 Lux. At the same time, it is conflicting at intensities 253 Lux and 405 Lux. Maximum is at 405 Lux. However a gap is there between values for EEAR between grade A & B at intensities 170 Lux, 310 Lux and 486 Lux. Maximum gap in value of EEAR from grade A to Grade B is for intensity at 486 Lux and minimum is for intensity at 170 Lux. EEAR for grade B & C are conflicting at all intensities including 310Lux. However, maximum is at 486 Lux and minimum is at 310 Lux. EEAR for grade C & D is conflicting at all intensities including 310 Lux. Maximum conflict is at 486 Lux and minimum is at 310 Lux. 102

10 These results are very clearly shown as graphs in Figure 4.4 for EEAR. Also shown in Table-4.4 is the distinction possibility between consecutive grades on the basis of EEAR values. Table-4.3 Range of Equivalent Ellipse Axes Ratio of Red Delicious Apple Intensity Equivalent Ellipse Axes Ratio based Model (Lux) Grade-A Grade-B Grade-C Grade-D Figure 4.4 Overlapping of Equivalent Ellipse Axes Ratio Range of Red Delicious Apple for Consecutive Grades at Different Intensities (Indicative Graphs, however, not to the exact scale) Table-4.4 Distinction Possibility Between Consecutive Grades on the Basis of Equivalent Ellipse Axes Ratio Intensity (Lux) Grades A & B Grades B & C Grades C & D 486 Yes No No 405 No No No 310 Yes No No 253 No No No 170 Yes No No 103

11 (c) Comparative Evaluation of HCF & EERA Based Fruit Shape Models In this section booth the models are compared and the results are tabulated in Table-4.5 in this regard. Table-4.5 indicates that on the basis of absolute success in HCF based Shape Model, there is small distinction possibility for grades A & B and as well as for grades B & C, whereas no distinction possibility is for grades C & D. Similarly, Table-4.5 also indicates that on the basis of absolute success in EEAR Based Fruit Shape model, there is small distinction possibility for grades A & B and there is nil distinction possibility for grade B & C as well as for grades C & D. From above analysis, it is inferred that both HCF & EEAR based Fruit Shape models are not consistent with regard to variation in fruit surface illumination. However, on the basis of distinction possibility, HCF based model is better than EEAR based model because both for grades B & C as well as for grade C & D the success counts are zero for EEAR based model. Table-4.5 Distinction possibility between consecutive grades on account of in Fruit Shape Models Grades Total combinations Success Heywood Circulatory Factor Based Model A & B B & C C & D Equivalent Ellipse Axes Ratio Based Model 104

12 4.3.2 Evaluation of Fruit Size Models (a) Perimeter Based Fruit Size Model As indicated in Table-4.6, with decrease in intensity from 486 Lux to 170 Lux, the value of Perimeter Range in pixels decreases for all the grades. Perimeter Range for grade A & B are conflicting at all intensities except 253 Lux. However, maximum is at 170 Lux and minimum is at 310 Lux. However a gap is there between values for Perimeter Range between grade A & B at intensity 253 Lux which is very high. Perimeter Range for grade B & C are conflicting at all intensities except 310 Lux. However, maximum is at 170 Lux and none at 310 Lux. Perimeter Range for grade C & D are conflicting at all intensities except 170 Lux. However, maximum is at 253 Lux and minimum is at 310 Lux. However a gap is there between values for Perimeter Range between grade C & D at intensity 170 Lux which is very high. Table-4.7 illustrates the extent of between different grades on the basis of perimeter range in pixels. Similarly as confirmed from Table-4.6 with decrease in intensity from 486 Lux to 170 Lux, the value of perimeter range in mm decreases in general. There are gaps present at same intensity as for value of perimeter range in pixels. Table-4.8 illustrates the extent of between different grades on the basis of perimeter range in mm. The results are very clearly shown as graphs in Figure 4.5 (a) and (b) for perimeter range in pixels and perimeter range in mm respectively. Also shown in Table- 4.9 is the distinct possibility between consecutive grades on the basis of perimeter range in pixels and perimeter in mm. 105

13 Table-4.6 Range of Perimeter (calibrated in pixels and mm) of Red Delicious Apple Intensity Grade-A Grade-B Grade-C Grade-D (Lux) Pixels mm Pixels mm Pixels Mm Pixels mm Intensity (Lux) Table-4.7 Overlapping between different grades on the basis of perimeter (pixels) Grades A & B Grades B & C Grades C & D 486 Yes Large Yes Large Yes Large 405 Yes Large Yes Large Yes Large 310 Yes Large No Nil Yes Small 253 No Nil Yes Large Yes Large 170 Yes Large Yes Large No Nil Table-4.8 Overlapping between different grades on the basis of perimeter (mm) Intensity Grades A & B Grades B & C Grades C & D (Lux) 486 Yes Small Yes Small Yes Small 405 Yes Small Yes Small Yes Negligible 310 Yes Negligible No Nil Yes Negligible 253 No Nil Yes Small Yes Small 170 Yes Small Yes Large No Nil 400 Range of Perimeter (calibrated in mm) of Red Delicious Apple A Grade A Grade B AGrade C Grade D Range of Perimeter (in mm) A B C A B C A B C B C D B C D D D D (a) Perimeter range calibrated in Pixels Illumination Intensity (lux) (b) Perimeter Range calibrated in mm Figure 4.5 Overlapping of Perimeter Range for Consecutive Grades at Different Intensities (Indicative Graphs, however, not to the exact scale) 106

14 Table-4.9 Distinction Possibility Between Consecutive Grades on the Basis of Perimeter Component Intensity Perimeter in Pixels Perimeter in mm (Lux) A & B B & C C & D A & B B & C C & D 486 No No No No No No 405 No No No No No No 310 No Yes No No Yes No 253 Yes No No Yes No No 170 No No Yes No No Yes (b) Hydraulic Radius Based Fruit Size Model As indicated in Table-4.10, with decrease in intensity from 486 Lux to 170 Lux, the value of Hydraulic Radius Range in pixels decreases for all the grades. Hydraulic Radius Range for grade A & B are conflicting at intensities 170 Lux and 253 Lux and no conflict at intensities 310 Lux, 405 Lux and 486 Lux. However, maximum is at 170 Lux. However a gap is there between values for Hydraulic Radius Range between grade A & B at intensities 310 Lux and 486 Lux. Maximum gap in value of Hydraulic Radius Range from grade A to Grade B is for intensity at 486 Lux and minimum is for intensity at 310 Lux. Hydraulic Radius Range for grade B & C are conflicting at intensities 170 Lux, 253 Lux, and 405 Lux and no conflict at intensities 310 Lux and 486 Lux. However, maximum is at 405 Lux. However a gap is there between values for Hydraulic Radius Range between grade B & C at intensities 310 Lux and 486 Lux. Maximum gap in value of Hydraulic Radius Range from grade B to Grade C is for intensity at 310 Lux and minimum is for intensity at 170 Lux. Hydraulic Radius Range for grade C & D are conflicting at all intensities except 405 Lux. However, maximum is at 486 Lux and minimum at 310 Lux. However a gap is there between values for Hydraulic Radius Range between grade C & D at intensity 405 Lux which is high.. Table-4.11 illustrates the extent of between different grades on the basis of Hydraulic Radius range in pixels. Similarly as 107

15 confirmed from Table-4.10 with decrease in intensity from 486 Lux to 170 Lux, the value of Hydraulic Radius range in mm decreases in general. Hydraulic Radius range in mm from grade A to Grade B is not conflicting at any intensity. From grade B to grade C it is conflicting at 405, 253 & 170 Lux and maximum conflict is at 405 Lux and minimum is at 253 & 170 Lux. However there are gaps present for 486 & 310 Lux, maximum at 486 Lux and minimum at 310 Lux. For grade C & D it is conflicting only at 170 Lux. Table-4.12 illustrates the extent of between different grades on the basis of Hydraulic Radius range in mm. Table-4.10 Hydraulic Radius Range (calibrated in pixels and mm) of Red Delicious Apple Intensity (Lux) Grade-A Grade-B Grade-C Grade-D Pixels Mm Pixels mm Pixels mm Pixels mm Table-4.11 Overlapping between different grades on the basis of Hydraulic Radius (pixels) Intensity Grades A & B Grades B & C Grades C & D (Lux) 486 No Nil No Nil Yes Negligible 405 No Nil Yes Small No Nil 310 No Nil No Nil Yes Negligible 253 Yes Negligible Yes Small Yes Small 170 Yes Small Yes Small Yes Small These results are very clearly shown as graphs in Figure 4.6 (a) and (b) for Hydraulic Radius range in pixels and Hydraulic Radius range in mm respectively. Also shown in Table-4.13 is the distinction possibility between consecutive grades. 108

16 Table-4.12 Overlapping between different grades on the basis of Hydraulic Radius Range (mm) Intensity Grades A & B Grades B & C Grades C & D (Lux) 486 No Nil No Nil No Nil 405 No Nil Yes Negligible No Nil 310 No Nil No Nil No Nil 253 No Nil Yes Negligible No Nil 170 No Nil Yes Negligible Yes Negligible (a) Hydraulic radius range calibrated in pixels (b) Hydraulic radius range calibrated in mm Figure 4.6 Overlapping of Hydraulic Radius Range for Consecutive Grades at Different Intensities (Indicative Graphs, however, not to the exact scale) Table-4.13 Distinction Possibility Between Consecutive Grades on the Basis of Hydraulic Radius Component Intensity Hydraulic Radius in Pixels Hydraulic Radius in mm (Lux) A & B B & C C & D A & B B & C C & D 486 Yes Yes No Yes Yes Yes 405 Yes No Yes Yes No Yes 310 Yes Yes No Yes Yes Yes 253 No No No Yes No Yes 170 No No No Yes No No 109

17 (c) Comparative Evaluation This section is dedicated to have a comparative analysis of perimeter and Hydraulic Radius based fruit size models for estimation of apple quality and the results are placed in Table-4.14 for this purpose. Table-4.14 indicates that on the basis of perimeter range in pixels and perimeter range in mm, there is negligible distinction possibility for all grade pairs. Table-4.14 Distinction possibility between consecutive grades on account of in Perimeter ranges in Fruit Size Model Grades Total Range of Value in Pixels Range of value in mm combinations Failure Success Failure Success A & B B & C C & D On the Basis of Perimeter Range Calibrated in Pixels: If distinction possibility for Perimeter based model at different intensities is examined, it is observed from Table-4.15 that on the basis of perimeter range value calibrated in pixels, following inferences are drawn: At 486 Lux, a nil distinction possibility is for all grades. At 405 Lux, a nil distinction possibility is for all grades. At 310 Lux, a nil distinction possibility is for grade A & B and for grade C & D, but at same intensity the distinction is possible for grade B & C. At 253 Lux the distinction is possible only for grade A & B but not for other grades. At 170 Lux, distinction is possible only for grade C & D not for other grades. 110

18 On the Basis of Perimeter Range Calibrated in mm: Similarly, if distinction possibility for Perimeter based model at different intensities is examined from Table-4.15, following inferences are drawn on the basis of perimeter range calibrated in mm: At 486 Lux, a nil distinction possibility is for all grades. At 405 Lux, distinction possibility is nil for all grade. At 310 Lux, a nil distinction possibility is for grade A & B and for grade C & D, but at same intensity the distinction is possible for grade B & C. At 253 Lux the distinction is possible only for grade A & B but not for other grades. At 170 Lux, distinction is possible only for grade C & D not for other grades. Table-4.15 Distinction Possibility Between Consecutive Grades on the Basis of Perimeter Component in Fruit Size Model Intensity Perimeter in Pixels Perimeter in mm (Lux) A & B B & C C & D A & B B & C C & D 486 No No No No No No 405 No No No No No No 310 No Yes No No Yes No 253 Yes No No Yes No No 170 No No Yes No No Yes After having complete comparative analysis as detailed above on the basis of results tabulated in Tables-4.14 and Table-4.15, it is revealed that perimeter based fruit size model is not consistent and this is a failure. On the basis of Hydraulic Radius (calibrated in pixels) Similarly, on the basis of Hydraulic Radius (calibrated in pixels) based fruit size model, following inferences are drawn from Table-4.16: There is large distinction possibility for grade A & B. 111

19 There is small distinction possibility for grade B & C. There is negligible distinction possibility for grade C & D. On the basis of Hydraulic Radius (calibrated in mm) Based on similar consideration, on the basis of Hydraulic Radius (calibrated in mm) based fruit size model, following inferences are drawn from Table-4.16: There is largest distinction possibility for grade A & B. There is small distinction possibility for grade B & C. There is large distinction possibility for grade C & D. Table-4.16 Distinction possibility between consecutive grades on account of in Hydraulic Radius ranges in Fruit Size Model Grades Total Range of Value in Pixels Range of value in mm combinations Failure Success Failure Success A & B B & C C & D To conclude the discussion, from the Table-4.16, it is revealed that Hydraulic Radius (calibrated in pixels) based fruit size model is better as compared to Hydraulic Radius (calibrated in mm) based fruit size model. Table-4.17 tabulates the distinction possibility between consecutive grades pairs on the basis of Hydraulic Radius based Fruit Size Model at different intensities. Hydraulic Radius Range (calibrated in pixels): From the results in Table-4.17, following inferences are drawn on the basis of distinction possibility with regard to Hydraulic Radius Range (calibrated in pixels): At 486 Lux, distinction is possible for grade B & C but not for other grades. At 405 Lux, distinction is possible for grade A & B and for grade C & D but not for other grades. 112

20 At 310 Lux, a nil distinction possibility is for grade C & D, but at same intensity the distinction is possible for grade A & B and for grade B & C. At 253 Lux, nil distinction possibility is for all grades. At 170 Lux, nil distinction possibility is for all grades. Hydraulic Radius (calibrated in mm): Similarly on the basis of Hydraulic Radius (calibrated in mm) Range, following inferences are made: At 486 Lux, distinction is possible for all grades. At 405 Lux, distinction possibility is nil for grade B & C, but at same intensity the distinction is possible for grade A & B and for grade C & D. At 310 Lux, distinction is possible for all the grades. At 253 Lux the distinction is not possible only for grade B & C but for other grades it is possible. At 170 Lux, distinction is possible only for grade A & B not for other grades. Table-4.17 Distinction Possibility Between Consecutive Grades on the Basis of Hydraulic Radius Component in Fruit Size Model Intensity Hydraulic Radius in Pixels Hydraulic Radius in mm (Lux) A & B B & C C & D A & B B & C C & D 486 Yes Yes No Yes Yes Yes 405 Yes No Yes Yes No Yes 310 Yes Yes No Yes Yes Yes 253 No No No Yes No Yes 170 No No No Yes No No To conclude the discussion, based on the results tabulated in Table-4.16 and Table-4.17, it is revealed that Hydraulic Radius (calibrated in mm) based model performs better. Table-4.18 tabulates the distinction possibility between consecutive grades on account of in Fruit Size Model. From Table-4.18, fruit size 113

21 models are compared and is revealed that Hydraulic Radius range model, whether calibrated in pixels or mm, distinction possibility is higher for all the grades pairs. Table-4.18 Distinction possibility between consecutive grades on account of in Fruit Size Model Grades Total combinations success in perimeter range model success in hydraulic radius range model Range of Value in Pixels Range of Value in mm Range of Value in Pixels A & B B & C C & D Range of Value in mm Evaluation of Bruise Size Models (a) Perimeter Based Bruise Size Model As indicated in Table-4.19, with decrease in intensity from 486 Lux to 170 Lux, the value of bruise perimeter in pixels decreases for all the grades. It has been established that for grades A and B, bruise perimeter is not at all intensities as the value of bruise size is zero for grade A. Bruise perimeter for grade B & C are conflicting at intensities 310 Lux, 405 Lux and 486 Lux and no conflict at intensities 170 Lux and 253 Lux. However, maximum is at 486 Lux and minimum at 310 Lux. However a gap is there between values for bruise perimeter between grade B & C at intensities 170 Lux and 253 Lux. Maximum gap in value of bruise perimeter from grade B to Grade C is for intensity at 170 Lux and minimum is for intensity at 253 Lux. Bruise perimeter for grade C & D are conflicting at all intensities except 253 Lux. However, maximum is at 405 Lux and minimum at 310 Lux. However a gap is there between values for bruise perimeter between grade C & D at intensity 253 Lux. Table-4.20 illustrates the extent of between different grades on the 114

22 basis of Bruise Perimeter in pixels. Similarly as confirmed from Table-4.19 with decrease in intensity from 486Lux to 170Lux, the value of Bruise Perimeter in mm decreases in general. There is no conflict in Bruise Perimeter in mm in any grade except for grade B & C at 486 and 405Lux. Table-4.21 illustrates the extent of between different grades on the basis of Hydraulic Radius range in mm. The results are very clearly shown as graphs in Figure 4.7 (a) and (b) for bruise perimeter range in pixels and bruise perimeter range in mm respectively. Also shown in Table-4.22 is the distinction possibility between consecutive grades on the basis of bruise perimeter range in pixels and bruise perimeter range in mm. Intensity (Lux) Table-4.19 Bruise Perimeter Range (calibrated in pixels and mm) of Red Delicious Apple Grade-A Grade-B Grade-C Grade-D Pixels mm Pixels mm Pixels mm Pixels mm Table-4.20 Overlapping between different grades on the basis of Bruise Perimeter (pixels) Intensit Grades A & B Grades B & C Grades C & D y (Lux) 486 No Nil Yes Small Yes Negligible 405 No Nil Yes Small Yes Small 310 No Nil Yes Negligible Yes Negligible 253 No Nil No Nil No Nil 170 No Nil No Nil Yes Negligible Intensity (Lux) Table-4.21 Overlapping between different grades on the basis of Bruise Perimeter (mm) Grades A & B Grades B & C Grades C & D 486 No Nil Yes Negligible No Nil 405 No Nil Yes Negligible No Nil 310 No Nil No Nil No Nil 253 No Nil No Nil No Nil 170 No Nil No Nil No Nil 115

23 (a) Perimeter range calibrated in pixels (b) Perimeter Range calibrated in mm Figure 4.7 Overlapping of Bruise Perimeter Range for Consecutive Grades at Different Intensities (Indicative Graphs, however, not to the exact scale) Table-4.22 Distinction Possibility Between Consecutive Grades on the Basis of Bruise Perimeter Component Intensity Bruise Perimeter in Pixels Bruise Perimeter in mm (Lux) A & B B & C C & D A & B B & C C & D 486 Yes No No Yes No Yes 405 Yes No No Yes No Yes 310 Yes No No Yes Yes Yes 253 Yes Yes Yes Yes Yes Yes 170 Yes Yes No Yes Yes Yes (b) Hydraulic Radius Based Bruise Size Model As indicated in Table-4.23, with decrease in intensity from 486 Lux to 170 Lux, the value of bruise hydraulic radius in pixels decreases for all the grades. It has been established that for grades A and B, bruise hydraulic radius is not at all intensities as the value of bruise size is zero for grade A. Bruise hydraulic radius for grade B & C are conflicting at intensities 310 Lux, 405 Lux and 486 Lux and no conflict at intensities 170 Lux and 253 Lux. However, maximum is at 486 Lux and minimum at 310 Lux. However a gap is there between values for bruise hydraulic radius between grade B & C at intensities 170 Lux and 253 Lux. Maximum gap in value of bruise hydraulic radius from grade B to Grade C is for intensity at

24 Lux and minimum is for intensity at 253 Lux. Bruise hydraulic radius for grade C & D are conflicting at all intensities except 253 Lux. However, maximum is at 405 Lux and no at 310 Lux. However a gap is there between values for bruise hydraulic radius between grade C & D at intensity 253 Lux. Table-4.24 illustrates the extent of between different grades on the basis of Bruise Hydraulic Radius in pixels. Similarly as confirmed from Table-4.23 with decrease in intensity from 486 Lux to 170 Lux, the value of Bruise Hydraulic Radius in mm decreases in general. There is conflict at all intensity in Bruise Hydraulic Radius in mm in all grades except for grade C & D at 486, 253 and 170 Lux. Table-4.25 illustrates the extent of between different grades on the basis of Bruise Hydraulic Radius in mm. The results are very clearly shown as graphs in Figures 4.8 (a) and (b) for Bruise Hydraulic Radius range in pixels and Bruise Hydraulic Radius range in mm Table-4.23 Bruise Hydraulic Radius Range (calibrated in pixels and mm) of Red Delicious Apple Intensity Grade-A Grade-B Grade-C Grade-D (Lux) Pixels mm Pixels Mm Pixels Mm Pixels Mm Table-4.24 Overlapping between different grades on the basis of Bruise Hydraulic Radius (pixels) Intensity Grades A & B Grades B & C Grades C & D (Lux) 486 No Nil Yes Negligible Yes Negligible 405 No Nil Yes Negligible Yes Negligible 310 No Nil Yes Negligible No Nil 253 No Nil No Nil No Nil 170 No Nil No Nil Yes Negligible 117

25 respectively. Also shown in Table-4.26 is the distinction possibility between consecutive grades on the basis of Bruise Hydraulic Radius range in pixels and Bruise Hydraulic Radius range in mm. Table-4.25 Overlapping between different grades on the basis of Bruise Hydraulic Radius (mm) Intensity Grades A & B Grades B & C Grades C & D (Lux) 486 No Nil Yes Negligible No Nil 405 No Nil Yes Negligible Yes Negligible 310 No Nil Yes Negligible Yes Negligible 253 No Nil No Nil No Nil 170 No Nil No Nil No Nil (a) Bruise Hydraulic radius range calibrated in pixels (b) Bruise Hydraulic radius Range calibrated in mm Figure 4.8 Overlapping of Bruise Hydraulic Radius Range for Consecutive Grades at Different Intensities (Indicative Graphs, however, not to the exact scale) Table-4.26 Distinction Possibility Between Consecutive Grades on the Basis of Bruise Hydraulic Radius Component Intensity Bruise Hydraulic Radius in Pixels Bruise Hydraulic Radius in mm (Lux) A & B B & C C & D A & B B & C C & D 486 Yes No No Yes No Yes 405 Yes No No Yes No No 310 Yes No Yes Yes No No 253 Yes Yes Yes Yes No Yes 170 Yes Yes No Yes No Yes 118

26 (c) Comparative Evaluation In this section, comparative evaluation of the Perimeter and Hydraulic Radius based Bruise Size models are compared so as to find the best alternative out of four choices. The study will also reveal to choose a combination of four different components emerged out of these two models for the estimation of apple quality, while considering the effect of fruit surface illumination as well. Table-4.27 tabulates distinction possibility between consecutive grades on account of in Bruise Perimeter ranges. From Table-4.27, it is revealed that on the basis of Bruise Perimeter range (calibrated in pixels), there is the largest distinction possibility for grades A & B and there is small distinction possibility for grades B & C. Moreover, there is negligible distinction possibility for grades C & D. Table-4.27 indicates that on the basis of Bruise Perimeter range (calibrated in mm), there is largest distinction possibility for grades A & B and there is small distinction possibility for grades B & C. Moreover, there is largest distinction possibility for grades C & D. From Table-4.27, it is revealed that Bruise Perimeter (calibrated in mm) based model is better as compared to Bruise perimeter (calibrated in pixels) based model. Table-4.27 Distinction possibility between consecutive grades on account of in Bruise Perimeter ranges Grades Total Range of Value in Pixels Range of value in mm combinations Failure Success Failure Success A & B B & C C & D If distinction possibility for Bruise perimeter based model at different intensities is examined from Table-4.28, it is inferred that: On the basis of Bruise perimeter (calibrated in pixels) 119

27 Following inferences are made on the basis of Bruise perimeter (calibrated in pixels) component: At 486 Lux, distinction is possible for grade A & B but not for other grades. At 405 Lux, distinction is possible for grade A & B but not for other grades. At 310 Lux, distinction is possible for grade A & B but not for other grades. At 253 Lux, distinction is possible for all grades. At 170 Lux, the distinction is not possible only for grade C & D but for other grades it is possible. On the basis of Bruise perimeter component (calibrated in mm) Following inferences are drawn on the basis of Bruise perimeter component (calibrated in mm): At 486 Lux, the distinction is not possible only for grade B & C but for other grades it is possible. At 405 Lux, the distinction is not possible only for grade B & C but for other grades it is possible. At 310 Lux, distinction is possible for all the grades. At 253 Lux, distinction is possible for all the grades. At 170Lux, distinction is possible for all the grades. Table-4.28 Distinction Possibility Between Consecutive Grades on the Basis of Bruise Perimeter Component Intensity Bruise Perimeter in Pixels Bruise Perimeter in mm (Lux) A & B B & C C & D A & B B & C C & D 486 Yes No No Yes No Yes 405 Yes No No Yes No Yes 310 Yes No No Yes Yes Yes 253 Yes Yes Yes Yes Yes Yes 170 Yes Yes No Yes Yes Yes 120

28 To summarize, from Table-4.27 and Table-4.28, it is revealed that Bruise perimeter (calibrated in mm) based model performs better and this model is also better than perimeter range model (calibrated in pixels) and Hydraulic Radius based model. On the basis of Bruise Hydraulic Radius (calibrated in pixels) On the basis of Bruise Hydraulic Radius (calibrated in pixels) based model, it is revealed from Table-4.29 that: There is largest distinction possibility for grade A & B. There is small distinction possibility for grade B & C. There is small distinction possibility for grade C & D. On the basis of Bruise Hydraulic Radius (calibrated in mm) It is analyzed from table-4.29 that: There is largest distinction possibility for grade A & B. There is small distinction possibility for grade B & C. There is small distinction possibility for grade C & D. From Table-4.29, it is revealed that Bruise Hydraulic Radius (calibrated in mm) based model is better. If distinction possibility is examined for Bruise Hydraulic Radius based model at different intensities from the Table-4.30, following inferences are drawn: On the basis of Bruise Hydraulic Radius (calibrated in pixels): Following inferences are drawn on the basis of bruise perimeter range calibrated in pixels: At 486 Lux, distinction is possible for grade A & B but not for other grades. At 405 Lux, distinction is possible for grade A & B but not for other grades. 121

29 At 310 Lux, the distinction is not possible only for grade B & C but for other grades it is possible. At 253 Lux, distinction is possible for all grades. At 170 Lux, the distinction is not possible only for grade C & D but for other grades it is possible. Table-4.29 Distinction possibility between consecutive grades on account of in Bruise Hydraulic Radius ranges Grades Total Range of Value in Pixels Range of value in mm combinations Failure Success Failure Success A & B B & C C & D Table-4.30 Distinction Possibility Between Consecutive Grades on the Basis of Bruise Hydraulic Radius Component Intensity Bruise Hydraulic Radius in Pixels Bruise Hydraulic Radius in mm (Lux) A & B B & C C & D A & B B & C C & D 486 Yes No No Yes No Yes 405 Yes No No Yes No No 310 Yes No Yes Yes No No 253 Yes Yes Yes Yes Yes Yes 170 Yes Yes No Yes Yes Yes On the basis of Bruise Hydraulic Radius (calibrated in mm), in mm: Following inferences are drawn on the basis of Bruise Hydraulic Radius calibrated At 486 Lux, the distinction is not possible only for grade B & C but for other grades it is possible. At 405 Lux, the distinction is possible only for grade A & B but not for other grades. 122

30 At 310 Lux, the distinction is possible only for grade A & B but not for other grades. At 253 Lux, the distinction is possible for all grades. At 170 Lux, the distinction is possible for all grades. From Table-4.29 and Table-4.30, it is concluded that Bruise Hydraulic Radius (calibrated in pixels) based model performs better and this model is also better than perimeter based model. From the Table-4.31, bruise size models are compared and analyzed that distinction possibility for both models is nearly equal but Bruise Perimeter based model is more consistent than Bruise Hydraulic radius based model. Table-4.31 Distinction possibility between consecutive grades on account of in Bruise Size Model Grades Total combinations success in Bruise Perimeter range model Range of Value in Pixels Range of Value in mm success in Bruise Hydraulic radius range model Range of Value in Pixels Range of Value in mm A & B B & C C & D Conclusion Based on the results of experimental validation attained as stated above, it is established that fruit surface illumination is one of the important quality determining parameters for estimation of fruit quality using external fruit attributes including fruit shape, size and bruise size. It has been established experimentally once again that it is not possible to have a clear cut distinction between different grades using fruit shape, fruit size and bruise size attributes, if fruit surface illumination is taken into account. In 123

31 fact, it is not possible to address this very complex issue using conventional image processing approach, where there is of different classes (grades). However, such type of complex multidimensional and multi modal problem can only be addressed by using artificial intelligence approach. This solution to this problem is provided in the next two chapters in the form of two alternate solutions. 124