应用改进遗传神经网络识别种蛋蛋形试验

针对人工检测种蛋蛋形劳动强度大，缺乏客观性，检测效率低，研究了自动快速、准确地识别鸡种蛋蛋形的方法。以蛋形指数和蛋径差为形状特征参数，利用机器视觉技术、矩技术和提出的改进遗传神经网络算法剔除畸形蛋。基于机器视觉和矩技术提取种蛋的长短径，剔除蛋形指数不合格种蛋后，再通过构建合理的遗传神经网络模型，以蛋径差作为神经网络输入参数，根据网络输出值识别种蛋蛋形。对过圆蛋、过尖蛋、畸形蛋和正常蛋检测准确率分别达到了97.10%、95.59%、94.87%和95.75%。研究种蛋蛋形自动识别方法对提高种蛋蛋形检测准确率和工作效率具有重要意义，试验结果表明提出的种蛋蛋形评价指标合理，用于识别种蛋正常蛋形，剔除畸形蛋准确率高，速度快，算法具有鲁棒性。

Experiment on automatic shape identification of hatching eggs based on improved genetic algorithm neural network

Shape inspection of hatching eggs is an important and hard work in farms, manual inspection lacks the objectivity and is time-consuming. In order to solve problems mentioned above, an automatic shape identification method was proposed based on machine vision, moment technique and improved genentic algorithm-neural network (GA-NN) algorithm. Egg shape index and radius differences were extracted as eggs shape feature parameters. An improved immune genentic algorithm was put forward to optimize topology structure of levenberg-marquardt back progagation-neural network (LMBP-NN). After egg shape index was identified , radius differences were used as inputs of LMBP-NN and its outputs were used to determine the hatching egg shape normal or not. The results indicated that the classification accuracy of this method reached 97.1% for longer eggs, 95.59% for shorter eggs, 94.87% for abnormal eggs and 95.75% for normal eggs' respectively. It is significant for shape identification of hatching eggs automatically, which can improve detection accuracy and efficiency. The neural network system for shape identification of hatching eggs has high accuracy and generalization ability, and the algorithm is feasible and robust.