Fast Weighing of Pistachio Nuts by Vibration Sensor Array

Transcription

1 International Journal of Electronics and Electrical Engineering Vol., No., August 0 Fast Weighing of Pistachio Nuts by Vibration Sensor Array Musa Ataş, Yahya Doğan, and İsa Ataş El-Cezeri Cybernetics & Robotics Laboratory, Siirt University, Siirt, Turkey Department of Computer Engineering, Siirt University, Siirt, Turkey Department of Electrical and Electronics Engineering, Dicle University, Diyarbakır, Turkey {hakmesyo, yahyadogan, isaatas@gmail.com Abstract Impact acoustic sound signal is previously used to discriminate open-shell pistachios from closed ones and for crack detection purposes. Weight of the pistachio samples can be utilized as a feature vector for sorting and grading processes. Nevertheless, traditional weighing procedure is time consuming. Moreover, efficient fast weighing system based on impact acoustic signals for pistachio nuts has not been studied yet. This study aims to discuss the design and evaluation of a real time fast weighing system for pistachio nuts. Proposed system can be extended to other agricultural or industrial products where weight information is critical as well. In order to eliminate the sensor noise and improve the signal quality, piezoelectric sensor arrays containing piezoelectric vibration sensors are employed. Final impact acoustic signal energy is determined by averaging the sensor array signals.0 pistachio samples with incremented weights ranging from 0. to. gr are utilized for calibration process of the sensor array. Extra two heavy objects (.0 and. gr) are participated to the calibration set also. In order to improve accuracy and achieve consistent measurements repetitive trials approach is adopted. Excessive repetition of experiments theoretically yields more accurate and consistent measurements with minimum standard deviation. Consequently it is observed that 0 times repetition scheme produces satisfactory results with % coefficient of variation and ms of computational cost indicates that proposed system can be applicable for fast weighing of pistachio nuts. TABLE I. U.S. STANDARD FOR GRADES OF PISTACHIO NUTS Size Designations Average number of nut per ounce Weight Range in gram Colossal Extra Large Large Medium Small < >0 >. gr.-. gr.-. gr gr <0. gr In order to classify pistachio nuts according to the weights specified in Table I, mechanical sieves have been established first. However, mechanical sieves/screening actually only able to classify the nuts according to their sizes. Although size of the pistachio nuts relates to the weight information, varied densities and inner structure of the kernels may also increase the misclassification rates. Fig. depicts certain pistachio nuts having different weights with similar sizes. Note that, theoretically mechanical sieve put them into single group, although in reality they belong to small, medium and large grade standards. Due to its prescribed deficiency, mechanical sieves are not applicable for this problem. As standards are directly related to the weight parameter, weighing system should be concerned. Index Terms fast weighing, impact acoustic, vibration sensor, pistachio sorting, piezoelectric sensor, sensor array I. INTRODUCTION Measurement is a vital element for all sorts of scientific researches and disciplines including engineering, manufacturing and production. Weighing is a type of measurement which we assess the objects that we deal with. Moreover, weighing is important for classification, sorting and grading issues because even weight parameter itself may be considered as a salient and discriminative feature. United States standards for grades of pistachio nuts. subsection, identifies the average weight of the nuts per ounce. Table I lists the weight limits of grading standards. Note that, gram conversion was made for the sake of clearness from the original document []. Figure. Almost similar size of pistachios having different weights left, middle, right, 0., 0.,. gram, respectively. Weight measurement can be handled by traditional or sensor based approaches such as balance scale, spring scale, strain gauge and impact acoustic sensors, respectively. Basic weighing approaches generally provide accurate results with high precision. But they are slow and integrating to the grading system is rather difficult. In order to address the problem fast and efficient weighing system is proposed in this work. Reference [] and [] proposed development of weigh in motion system using acoustic emission sensors. Reference [] also dealt with wireless capability of the system and both of them tried to estimate weights of Manuscript received May, 0; revised October, 0. 0 Int. J. Electron. Electr. Eng. doi: 0./ijeee...-

2 International Journal of Electronics and Electrical Engineering Vol., No., August 0 trucks based on road vibration signals. Piezoelectric vibration sensors can be used for detection of impact acoustic signals. In the literature various vibration sensor based systems have been studied. Pearson et al. showed that by building microphone, Digital Signal Processing (DSP) device and air rejection nozzle separator system, almost % classification accuracy for open-shell and closed shell pistachio nuts can be achieved []. Other studies related to the impact acoustic can be read []-[]. For image based sorting systems, Haff et al. studied sorting of in-shell pistachios from kernels using color images and achieved.% accuracy for regular in-shell pistachio from kernels. However for smaller in-shell pistachio this accuracy rate drops to % and % for Discriminant Analysis (DA) and K-Nearest Neighbor (KNN) approaches, respectively []. Ghazanfari et al. used Fourier descriptors and MLP as features and a classifier for grading the pistachios into three United States Department of Agriculture (USDA) size grades and closed-shell class, respectively. They achieved.% overall classification accuracy []. Another image based study was conducted by Kouchakzadeh and Adel for discriminating five different varieties of pistachios and obtained.% accuracy rate [0]. It should be noted that both studies performed the classification process in an off-line manner and generated image dataset was actually made up of ideal pistachio postures and positions. Thus for a real time operation classification performance might be adversely affected due to the challenging cases that may be arisen from pistachio nuts positions while dropping. The objective of this study is to assess the feasibility and the efficiency of the impact acoustic based pistachio sorting system that aims to grade pistachios by using Vibration Sensor Array (VSA) in the real-time manner. Section II describes detailed information about impact acoustic signal generation, major components of the proposed system and feature extraction methods. Calibration process, experimental results and discussions are presented in Section III. Consequently, a couple of concluding remarks and future projections are drawn at final section. similar velocity and can be considered as constant and therefore velocity difference among different samples can be ignored. Hence, we can say that P is directly proportional to the mass of the falling object. Therefore it is reasonable to use impact energy to measure the mass of the falling object. There exist several types of vibration sensors including piezoelectric accelerometer, velocity sensor, proximity probes and laser displacement sensors []. Due to its low price, small size and convenient to integrate to setup, piezoelectric vibration sensor MEAS is preferred. Fig. illustrates MEAS vibration sensor. Figure. A horizontal type the MiniSense 00 from measurement specialties piezoelectric vibration sensor. Basically sensor produces small AC and large voltage (up to ±0V) when the film, piezoelectric element, is hit. It is sensitive enough to capture any small impacts and can be used for a flexible switch as well. MΩ resistor should be wired to down voltage to the Analog Digital Converter (ADC) levels. B. Architecture and Major Components of Proposed System Proposed system consists of vibration sensors, MΩ resistors, Arduino UNO electronic cards, a plexiglass pipe with 0 cm length and cm diameter and a computer. Here Arduino UNO is preferred because beyond it has low price, it supports analog inputs as well. Besides, its ADC sampling frequency rate (00HZ) is higher than the sensor output frequency (0HZ) which makes it convenient in terms of Nyquist theorem. VSA module is made up of five sensors. Fig. demonstrates the developed VSA module. II. MATERIAL AND METHODS A. Impact Acoustic Signal For achieving maximum throughput, pistachio nuts are released from the elevated position as a free fall movement under the gravitational force. Systems that use conveyor belt are usually slower than the aforementioned approach. When an object hits the material, impact acoustic sound signals are propagated. Previous studies []-[] employed those sound signals. Another alternative is using vibration sensor to acquire impact/hit energy. Impact energy is proportional to the momentum. As () suggests with nearly constant velocity mass parameter would be discriminative. P m v () here P denotes momentum, m is the mass of the object and v designates velocity. Almost all samples have Figure. A typical VSA module installed on the tube. As it is seen from the Fig., vibration sensors are soldered with silicon on the ring so that they can catch the dropping objects inside the plexi-glass tube. Three VSA modules are positioned at certain altitude on the pipe. Approximate distances between VSA modules are cm in general. Each VSA module is wired to the specific 0 Int. J. Electron. Electr. Eng.

3 International Journal of Electronics and Electrical Engineering Vol., No., August 0 Arduino UNO card. In this way, real-time parallel processing can be handled by the simple Arduino code. Code snippets of the Arduino are available below. int thr=0; int limit=0; float t=0; void setup() { Serial.begin(00); void loop() { int s=analogread(a0); int s=analogread(a); int s=analogread(a); int s=analogread(a); int s=analogread(a); int signal=(s+s+s+s+s)/; if (signal_>thr) { for (int i=0;i<limit;i++){ signal =(analogread(a0)+analogread(a)+ analogread(a)+analogread(a)+ analogread(a))/; t+= signal; Serial.println(t); t=0; that we utilize it as a feature vector in this study. Please note that, there is no curse of dimensionality problem here because only single input simplifies the developed algorithm along with the system can be accounted as a satisfactory confidence level. III. A. Calibration Process EXPERIMENTAL RESULTS In order to produce consistent and reliable results calibration process should be carried out. Developed system may produce different results according to the shape of the object and dropping conditions. That is, if object is very small in size it may escape the VSA module which may yields false reading. Similarly sometimes object hit some sensors directly and remaining are affected weakly from this impact. As a result there may be slight divergence from the ideal read. In order to address this particular problem we repeat the process and then average them. In this study, we investigate the optimum number of repetition under the consideration of processing speed. To do that, 0 pistachio samples with incremented weights range between 0. to. gr are utilized for building calibration set. Extra two heavy objects (.0 and. gr) are participated to the calibration set also. Fig. depicts those samples. Figure. Captured impact signals from the VSA module as a time series data. Similarly, Fig. shows digitized impact energy of hit samples to the VSA module. Note that, Arduino UNO has 0 bit ADC and can produce maximum 0-0 positive values. With VSA module, sensor noise is suppressed by averaging the signal as well. As () indicates, total amount of energy is utilized as a feature in this study. 0 i, j, k k () i j TE here, TE and S denote the total energy and signal value (amplitude) of ADC, respectively. For each VSA module, signals of sensors are averaged and then summed up. As we have three modules on the system, energies of VSA modules are averaged to get the overall impact energy S Figure. Samples used in calibration process. Each calibration sample is employed for one repetition to ten repetition incrementally. Mean and standard deviation are calculated to obtain percent coefficient of variation (%CV) that we think it is more representative than other statistics. Equation () shows the percent coefficient of variation formula. % CV 00 () Table II demonstrates influence of the repetition on the sensor measurement with respect to the percent coefficient of correlation values. Similarly, improvement on measurement can also be seen Fig. and Fig. as each calibration samples and mean value, respectively. 0 Int. J. Electron. Electr. Eng.

4 International Journal of Electronics and Electrical Engineering Vol., No., August 0 TABLE II. EFFECT OF REPETITION ON %CV x x x x x x x x x 0x 0, gr 0, gr 0, gr 0,0 gr 0,0 gr 0 0, gr, gr 0,0 gr 0, gr, gr,0 gr, gr Average were participated to the calibration set also. In order to improve accuracy and achieve consistent measurements repetitive trials scheme was adopted. Experiments revealed that 0 times repetition scheme produces satisfactory results with % coefficient of variation and ms of computational cost indicates that proposed system can be applicable for fast weighing of pistachio nuts. In the future, it is intended to predict weight correction factor of the proposed system for weighing in gram scale. Detailed test will be performed to determine generalized performance of the developed system. ACKNOWLEDGMENT This study was funded by the Scientific and Technological Research Council of Turkey (TÜBİTAK) under grant no. E0. Special thanks to the El-Cezeri laboratory stuff Muhammed Said Ataş for his valuable efforts on conducting repetitive and exhaustive experiments. REFERENCES U.S.D.A., U.S. standards for grades of pistachio nuts in the shell, Technical Report, 00. [] J. M. Bowie, Development of a weigh-in-motion system using acoustic emission sensors, Ph.D. dissertation, Dept. Civil, Env. and Constr. Eng., Univ. of Central Florida, Orlando Florida, 0. [] R. Bajwa, Wireless weigh-in-motion: using road vibrations to estimate truck weights, Ph.D. dissertation, Dept. Elect. Eng. and Computer Science, Univ. of California, Berkeley, 0. [] T. C. Pearson, Detection of pistachio nuts with closed shells using impact acoustics, Applied Engineering in Agriculture, vol., no., pp. -, 00. [] A. E. Cetin, T. C. Pearson, and A. H. Tewfik, Classification of closed- and open-shell pistachio nuts using voice-recognition technology, Transactions American Society of Agricultural Engineers, vol., no., pp. -, 00. [] H. Kalkan, N. F. Ince, A. H. Tewfik, Y. Yardimci, and T. C. Pearson, Classification of hazelnut kernels by using impact acoustic time-frequency patterns, EURASIP Journal on Advances in Signal Processing, vol. 00, January 00. [] T. Pearson and N. Toyofuku, Automated sorting of pistachio nuts with closed shells, Applied Engineering in Agriculture, vol., no., pp. -, 000. [] R. P. Haff, T. C. Pearson, and N. Toyofuku, Sorting of in-shell pistachio nuts from kernels using color imaging, Applied Engineering in Agriculture, vol., no., pp. -, 00. [] A. Ghazanfari, J. Irudayaraj, A. Kusalik, and M. Romaniuk, Machine vision grading of pistachio nuts using Fourier descriptors, Journal of Agricultural Engineering Research, vol., no., pp. -,. [0] A. Kouchakzadeh and B. Adel, Discrimination of pistachios varieties with neural network using some physical characteristic, International Journal of Emerging Sciences, vol., no., pp., 0. [] H. N. Norton, Handbook of Transducer, Prentice Hall,, ch. -. [] Figure. Repetition vs %CV for each calibration sample. Figure. Average repetition vs %CV line. Table II, Fig. and Fig. show that as number of repetition increase, mean value of sensor arrays are become more consistent and reliable because standard deviation and %CV value become smaller and smaller. IV. CONCLUSIONS Main objective of this study is to discuss the design and evaluation of a real time fast weighing system for pistachio nuts. Proposed system can be extended to other agricultural or industrial products as well. In order to achieve better Signal to Noise Ratio (SNR), piezoelectric sensor arrays containing piezoelectric vibration sensors were utilized. Total impact acoustic signal energy was determined by averaging the sensor array signals. 0 pistachio samples with incremented weights ranging from 0. to. gr were utilized for calibration process of the sensor array. Extra two heavy objects (.0 and. gr) 0 Int. J. Electron. Electr. Eng. Musa Ataş is an Assistant Professor in the Department of Computer Engineering at the University of Siirt where he has been a faculty member since 0. He is a founder and principal coordinator of the El-Cezeri Cybernetics and Robotic Laboratory. Musa completed his undergraduate, MS and Ph.D. at Middle East Technical University/ Turkey. His research interests lie in the area of artificial intelligence, autonomous systems, machine and computer vision, machine learning, robotics, virtual reality

5 International Journal of Electronics and Electrical Engineering Vol., No., August 0 and programming languages specifically domain specific languages as Open Cezeri Library framework ranging from theory to design to implementation, with a focus on improving software quality. In recent years, he has focused on machine vision systems and impact acoustic. He has collaborated actively with researchers in several other disciplines of computer science, agricultural and food science. Currently, he conducts two projects, classification of pistachio nuts by machine vision and aflatoxin detection in pistachio nuts by hyper spectral imaging and machine vision, respectively. Musa has served on roughly ten conference and workshop program Yahya Doğan works as an assistant and he is a graduated student in the Department of Computer Engineering at the University of Siirt where he has been a faculty member since 0. He is a co-founder and stuff of the El-Cezeri Cybernetics and Robotic Laboratory. Yahya completed his undergraduate at Sakarya University. Currently his MS is at Fırat University, Turkey. His research interests lie in the area of machine vision, machine learning. In recent years, he has focused on industrial cameras. His MS thesis is related to the prediction of ideal exposure time of industrial cameras for multispectral/hyper spectral imaging and machine vision. Yahya has served on roughly five conference and workshop program İsa Ataş works as a lecturer and he is a Ph.D. student in the Department of Electrical and Electronics Engineering at the University of Dicle where he has been a faculty member since 00. İsa completed his undergraduate and MS at Dicle University with Electrical and Electronics Engineering. Currently his Ph.D. is at Dicle University / Turkey too. His research interests lie in the area of machine learning and patch antenna design. In recent years, he has focused on design of patch antenna. His Ph.D. thesis is related to the design and implementation of high gain aperture coupled microstrip patch antenna. İsa has served on roughly five conference and workshop program 0 Int. J. Electron. Electr. Eng.