Research On Recognition Methods Of Egg Exterior Quality And Experiment Device

The experiment device is designed for detecting egg exterior quality, appropriate light source and background are selected. The captured images are pre-processed with the methods of declining noise,image enhancement,binarization,edge detection, etc. Iterative method is adopted to obtain the best segmentation threshold while image binarization, and optimum-Ostu algorithm is put forward for boundary extraction. For dirty,crack images segmentation, firstly Ostu algorithm is combined with sobel operator, as a result internal information is supplemented on the basis of guaranteeing clear image edge, and then threshold recognition method is used for getting rid of interferential value and lines, the accuracy of extracting the fault feature is elevated, at last eggs are detected according to the number of white pixels occupied by faults.Extracting feature parameters of apparent quality,area of egg image is used to evaluating egg weight, the correlation coefficient is 0.9922. Egg form index is selected as shape measurement, when calculating egg form index, modified minimum rectangle is proposed for calculating egg long and short axes, the axes and actual measurement values are correlate with 0.9733,0.9797. Based on analysis of comparing the standard BP and three kinds of modified BP algorithms, the strategy is advocated that firstly neural network weights are optimized through GA algorithm, and then training BP network by L-M algorithm, network training conditions are improved validly. Network simulation is carried on with examination samples, the classification accuracy of GA-LM network is 93.8% for egg weight, and 100% for egg shape.