基于L-M算法优化BP神经网络的储粮害虫分类识别研究

Research on the Detection and Classification in Stored-grain Based on BP Neural Network with L-M Optimization

以储粮害虫为对象,研究了利用数字图像处理技术与BP神经网络技术实现离线检测与分类识别。首先对4类常见储粮害虫进行图像采集、预处理以及9个常用形态学特征的提取,再通过特征分析把有效特征压缩至6维,将其作为BP神经网络的输入参数,对应的储粮害虫的类别代号作为输出参数,构造BP神经网络,并在网络训练过程中利用L-M算法进行优化。最后通过实验证明该方法在害虫识别算法中稳定性好,收敛速度快,预测精度高。

Taking the stored-grain pest as an object,it presents a method of offline detection and classification based on the technology of digital image processing and optimized BP neural network.Through comparison and analysis of the morphological feature parameters of pest outline,it reduces the 9 effective features to 6,and uses these features as input parameters of BP neural network.It establishes the neural network model between the feature parameters and pest categories,and improves the model with the numerical optimization method of L-M.The experiment results show the method proposed has obvious advantages of good stability,fast convergence and high accuracy.