基于改进Hu矩和遗传神经网络的稻飞虱识别系统

Recognition System of Rice Planthopper Based on Improved Hu Moment and Genetic Algorithm Optimized BP Neural Network

摘要:

针对稻飞虱识别实时性差和BP神经网络分类有一定误差的问题，设计了一种基于DSP硬件平台和遗传神经网络算法的稻飞虱识别系统。系统硬件以AT89S52单片机控制拍摄移动装置，以DM6437处理器作为算法处理平台；系统软件设计主要包括基于改进Hu矩的特征值提取和基于遗传算法优化神经网络的识别算法。系统通过CCD摄像机拍摄稻飞虱视频信号传送到DSP识别系统，从中提取图像，识别图像中的稻飞虱。实验对稻飞虱、水蝇和潜蝇等80个样本进行了训练和测试，结果表明遗传神经网络对稻飞虱的正确识别率达到90%。

Abstract:

For the problems of poor real-time of rice planthopper recognition and a certain error of BP neural network classifier, a rice planthopper recognition system was designed based on DSP hardware system and genetic algorithm optimized BP neural network. AT89S52 microcontroller was used to control the mobile device. DM6437 was used as processing platform. Mathematical morphology algorithm, improved Hu moment, and genetic algorithm optimized BP neural network algorithm were used for segmentation. The video camera was used to shoot crop video. Then, the video signal images were transformed to the DSP recognition system. The rice planthopper could be identified from these images. The experiment was carried out on 80 samples, including rice planthopper, ephydrid and miner. Results showed that the accuracy of genetic algorithm optimized BP neural network reached to 90%.