基于SVM和CNN组合模型的黄瓜病斑叶片检测与识别

Detection and Recognition of Diseased Cucumber Leaves Based on SVM and CNN Combined Model

为建立准确高效的黄瓜病斑叶片的检测与识别算法提供参考,针对黄瓜叶片常见病斑检测与识别时存在的环境适应性差、识别精度低等问题,提出基于改进的SVM(支持向量机)和CNN(卷积神经网络)组合模型的黄瓜病斑叶片检测与识别算法。依据黄瓜设施场景特征,首先对病斑图像进行色彩增强,通过直方图均衡化对图像进行再处理,利用优化的HOG+SVM分类器对黄瓜叶片进行提取;通过稀疏滤波器及增加偏置对CNN算法进行改进,识别出叶片的病斑类别。结果表明:在黄瓜设施场景下,改进SVM和CNN组合模型的黄瓜病斑叶片检测与识别算法对叶片提取的查准率及差全率分别达87.21%和88.77%,对病斑的整体识别精准率为91.9%。算法实时性强,具有实际推广应用前景。

In order to provide references for establishing an accurate and efficient detection and recognition algorithm for cucumber diseased leaves,aiming at the problems of poor environmental adaptability and low recognition accuracy in the detection and recognition of common disease spots on cucumber leaves,an algorithm for detection and recognition of disease spots on cucumber leaves based on the combination model of SVM and CNN was proposed.According to the scene characteristics of cucumber facilities,the color of the lesion image was enhanced firstly,the image was reprocessed by histogram equalization,and the leaves of cucumber were extracted by HOG+SVM classifier;the CNN algorithm was improved by sparse filter and adding bias to identify the lesion types of leaves.The experimental results showed that in the cucumber facility scenario,the accuracy and total difference of leaf extraction could reach 87.21%and 88.77%respectively,and the accuracy of the whole identification of disease spots was91.9%.At the same time,the algorithm had strong real-time performance and had practical application prospects.