基于人工神经网络的田间秸秆覆盖率检测系统

Measuring System for Residue Cover Rate in Field Based on BP Neural Network

以VC＋＋为工具，田间实拍图像为研究对象，在分析田间秸秆和土壤纹理特征差别的基础上，设计了BP神经网络秸秆覆盖率检测系统。该系统采用了神经网络与纹理特征相结合的方法提取秸秆，并以纹理特征熵值为标准建立了网络输入层学习样本选取准则。人工模拟和田间试验表明，设计的BP神经网络秸秆覆盖率检测系统对田间秸秆的识别率达90％以上，秸秆覆盖率计算误差可控制在5％以内；与传统的拉绳法相比，检测效率提高50～120倍。

According to the analyses of the texture differences between straw and soil, a new BP neural network measuring system for residue cover rate is designed. By taking the filed photos as the research objectives, this system was developed through VC ＋ ＋ programming tools. Straws were detected by combining the texture features and BP neural network. Selection standard of learning samples for input nodes was constructed based on the entropy in the system. Artificial simulation and field testing indicated that the new measuring system could detect over 90 % of the straws in the field and control the counting error of residue cover rate under 5 %. Compared with the traditional manual measuring, the measuring efficiency in the new system could be improved by 50--120 times.