Study On Automatic Identifying Quality And Fertility Of Hatching Eggs Based On Machine Vision System

Identifying quality and fertility of hatching eggs are an important and hard work in the farms. Manual inspection suffers from visual stress and tiredness and is low accuracy and time-consuming. An automatic and practical detection system based on machine vision system and ANN is developed instead of manual inspection of hatching egg for improving detecting accuracy and effciency.1. The machine vision hardware system is built for identifying exterior quality and fertility of hatching egg .The light source and background color are found out through a lot of experiments. Camera calibration is done for correcting image distortion, and its accuracy is able to match the demand of identifying exterior quality of hatching egg.2. Based on machine vision technique, criterion is proposed for comprehensive evaluating egg's exterior quality by weight, shape, eggshell defect feature and eggshell color, and method of egg quality classification is developed.(1) The projection area of egg image is extracted by 0-order moment and used to classify egg weight instead of metage. The classification accuracy is 97.73% for bigger eggs, 97.04% for normal eggs, and 96.51% for smaller eggs.(2) Threshold recognition and 8-connected boundary tracking method are combined to extract the defect feature on eggshell, and its classification accuracy is 91.25% for cracked eggs, 94.18% for dirt stained, blood spotted eggs and 96.36% for normal eggs.(3) Egg shape index and radius differences are extracted as shape feature parameters, a two-step shape measurement method is proposed based on machine vision, moment technique and neural network. An improved immune GA algorithm is put forward, which is used to optimize topology structure of LMBP neural network for detecting quality of hatching egg automatically. After identified egg shape index, radius differences are used to get rid of abnormal eggs. The classification accuracy is 97.1% for longer eggs, 95.59% for shorter eggs, 94.87% for abnormal eggs and 95.75% for normal eggs.(4) Eggshell color is identified by the improved immune GA-NN with the feature of hue frequency value extracted, and the classification accuracy is 95.6% for lighter color eggshell, 95.8% for normal color eggshell and 91.3% for darker color eggshell.3. An improved SA-PSO is put forward, which is used to optimize topology structure of BP neural network for detecting fertility of hatching egg automatically. Detection method of fertility of hatching egg during the entire hatching period is developed systematically. Hue frequency value of the hatching egg image is selected as the input of neural network. Fertility of hatching egg is identified by the improved PSO neural network.The neural network system for fertility of hatching eggs detection has a high accuracy and generalization ability. The classification accuracy is 92.5%,98.3% and 100% for early , middle and later hatching period of eggs.4. A software package for all functions is integrated finally which lay a foundation for further study.