Egg Quality Non-Destructive Detection Based On Multi-Sensor Fusion

Non-destructive detection with single sensor has good excitation-response characteristic on general quality index of agricultural products, but relative low detection precision and detection stability existed in the single sensor detection. Furthermore, some complex and subdividing quality index of agricultural products can not be detected by single sensor which showed the detection scope of single sensor has some blind area. On the other hand, the detection on agricultural products tends to relate to the improvement of export standards. For non-destructive detection, sometimes comprehensive quality information are required. Consequently, multi-sensor fusion is necessary for non-destructive detection on agricultural products from the aspects of application. The sensor information pre-processing, the sensors fusion on data level, the sensors fusion on feature layer level, the expert system model and the virtual instrument technology have been used in egg quality non-destructive detection. Comparing with the non-destructive measurement by single sensor, the detection precision, the detection stability and the detection scope can be improved effectively by using multi sensor fusion technology. Meanwhile, some complex quality parameters of eggs during storage and transportation which can not be analyzed by single sensor can be detected. Based on these studies, application software and hardware system could be designed which includes an expert system and a virtual instrument system. The specific research works are as follows:1. Pre-processing of the sensor information and the egg quality index determination methods suitable for multi-sensor fusion detectionThe Fast-independent component correlation algorithm (Fast-ICA) on the basis of fixed point iteration method was used in output information de-noising and optimization of machine vision sensor. The wavelets transform algorithm on the basis of lifting scheme method was used in output information de-noising and optimization of acoustics (impact excitation) sensor. Both of the two methods have achieved good de-noising results. Then, the suitable experiment conditions for egg of electronic nose have been determined by multiple tests. The pattern recognition algorithm has been studied systematically for electronic nose outputs information aiming at egg odor. The PNN (probabilistic neural network) was selected as the best pattern recognition algorithm for egg detection from 8 algorithms by algorithm performance tests. The tests showed that PNN has high recognition stability, good recognition precision and fast running speed. The suitable electronic nose sensor groups for egg crack (sensor 2,7 and 8) detection and egg freshness (sensor 2,5 and 8) detection were determined respectively by sensor loading analysis and the regression analysis.2. The non-destructive detection of egg freshness based on sensor fusion at the data levelThe Hafu unit and Total Volatile Basic Nitrogen (TVB-N) of eggs were detected by machine vision and electronic nose respectively on the data fusion level. A data fusion method which distributes weight of the characteristic parameters by the contribution degree of forecasting the target parameters was proposed. The data fusion methods used to determine the characteristic parameters and analyze model construction process by machine vision and electronic nose were proposed respectively, which solves the problem of information absence and interior noise existed in general non-destructive detection procedure.The verification study showed that sensor data fusion enhanced the precision, stability and portability of the Hafu unit and TVB-N forecast models in eggs. On this basis, the suitable value of the support vector regression (SVR) input parameters, the SVR type, the SVR kernel function and related SVR interior structural parameters were determined by simulation tests and practical verification.Then, a new hybrid method to determine the SVR interior structural parameters on the basis of particle swarm optimizer (PSO) algorithm combines with the grid searching method was designed which fully develops the high searching speed of PSO and with good stability of the grid searching. The verification results showed that the new hybrid method remains with the good searching accuracy as the general grid searching method while the searching speed is 4 times faster than the general grid searching method. The method could meet the requirement of online detection with medium speed.3. The non-destructive detection of egg crack and related exterior quality based on multi-sensor fusion at the characteristic levelFirstly, the cracks of eggs were divided into two levels which consisted of the size of crack and the depth of crack. On this basis, the multi-sensor fusion detection at egg crack (including the size and depth of crack) characteristic level was carried out by the machine vision sensor, the acoustic sensor and odor sensor (E-nose). The egg crack systemic detection with SVM model was proposed by 3 sensors fusion for the first time. The model performance parameters and verification results showed the model had credible structure and good identification ability on eggs crack detection. The system analysis results could be used for the sorting process before egg processing.Then, the multi-sensor fusion detection in egg shell intensity characteristic level was studied by the machine vision sensor and the acoustic sensor. The identification threshold value of egg shell intensity testability and related identification model were determined by the variation coefficient test of characteristics frequency in impact excitation and the measurement of shell intensity in crack eggs. Furthermore, sensor fusion method was taken to determine set of characteristic parameters. Finally, the egg shell intensity systemic prediction (multiple linear regressions, MLR) model on the basis of backward elimination was established. The verification results showed the model obtained high prediction accuracy (the average prediction error was below 5%) and good stability (the standard deviation of the prediction error was 0.003) which solves the problem of which traditional non-destructive detection could not detect egg intensity with different types (eggs with crack and without crack) simultaneously. The research could offer the key detection data for egg processing.4. The non-destructive detection of egg freshness based on multi-sensor fusion at the characteristic levelThe multi-sensor fusion of machine vision and E-nose on egg freshness detection was studied in the characteristic level on the basis of Dempster-Shafer (D-S) evidence theory. Then, the egg freshness prediction model on the basis of multi-sensor fusion was established. According to the results of main factor analysis, the characteristic parameters (the average value of Hue in egg content image and the G/G02 value of E-nose) with highest prediction contribution rate on egg freshness were selected. Meanwhile, a group of criterion rules for egg freshness detection based on D-S evidence were proposed. Furthermore, a new basic probability assignment (BPA) method was designed based on SVM classification method. The comparative results showed that the egg freshness perdition model optimized by SVM classification method could fully utilize the adaptive ability of SVM and the uncertainty deduction ability of D-S evidence theory. The sensor fusion could decrease the basic probability assignment of indeterminacy (U is less than 0.0009) substantially which solves the identification blind area and low stability in single sensor non-destructive detection. The sensor fusion model also had the advantages of high running speed and good transferability.Comparing to the single sensor model prediction results, the machine vision-electronic nose fusion model highly enhanced the prediction accuracy and stability on egg freshness detection (the prediction accuracy of sensor fusion model were better than the prediction model using parameters from machine vision and E-nose independently, and the standard deviation the prediction accuracy of sensor fusion model were 47.69% and 27.23% of the prediction model using parameters from machine vision and E-nose independently).5. The design and realization of egg comprehensive quality non-destructive detection expert system based on multi sensor fusionThe egg comprehensive quality non-destructive detection knowledge depository and management system on the basis of multi sensor fusion was designed by refining and analyzing the physiology property, storage characteristics of eggs and the related non-destructive detecting stimulus-response characteristics. Furthermore, the system analysis principle and the paradigm theory was used to establish the egg comprehensive quality non-destructive detection database, model-base and related management system.By using the D-S evidence theory prediction model, SVM identification model and SVR prediction model as the kernel, the egg non-destructive detection inference engine on the basis of multi sensor fusion was designed in heuristic structure. On this basis, the rule base and related management system of new model construction in the egg non-destructive detection were studied and designed. Then, we integrated the contents above by coordination mechanism to design the component-based egg comprehensive quality non-destructive detection expert system on the basis of multi sensor fusion (MSEAES) for the first time.The MSEAES realizes the modules of prediction and identification, data management, new model establishment, dynamic simulation, detection strategy selection and system help documentary which have comprehensive function and good decision-making ability. The research of MSEAES offered key data and application module for establishing and developing agricultural products non-destructive detection system platform.6. The egg comprehensive quality non-destructive detection and related application hardware platform realizationBased on the multi-sensor fusion non-destructive detection research on egg quality and the important standards for egg storage in China, a comprehensive quality non-destructive detection flow which was suitable for most egg detection requirements was proposed. Meanwhile, an egg comprehensive quality systemic criterion was designed for the research which consists of the crack, the freshness, the spot and the verification accuracy in egg detection. Finally, the virtual instrument (â…¥) technology was used in the egg non-destructive detection procedure on the basis of multi-sensor fusion and a comprehensive egg quality detection system was designed. The egg comprehensive quality detection system was suggested based on the MSEAES in part 6. The system running tests showed that the system could detect and grade eggs automatically according to the quality with high intelligent degree. The static grading test, the dynamic grading test and the continuous dynamic test of the system showed that the egg comprehensive quality detection system could achieve good accuracy (the detection accuracy in dynamic test was 90%) and low missing rate (the missing rate in dynamic test was less than 4%) at a system running speed of 53 eggs per minute.