基于改进VGG卷积神经网络的棉花病害识别模型

为实现自然条件下棉花病害图像准确分类,提出基于改进VGG-16卷积神经网络的病害识别模型。该模型在VGG-16网络模型基础上,优化全连接层层数,并用6标签SoftMax分类器替换原有VGG-16网络中的SoftMax分类器,优化了模型结构和参数,通过微型迁移学习共享预训练模型中卷积层与池化层的权值参数。从构建的棉花病害图像库中随机抽取病害图像样本作为训练集和测试集,用以测试该方法的性能。试验结果表明:该模型能有效提取出棉花病害叶片图像的多层特征图像,并通过Relu激活函数的处理更能凸显棉花病害的边缘信息与纹理信息,分辨率为512像素×512像素图像在样本训练与验证试验效果最好。在平均识别准确率方面,本研究模型较BP神经网络、支持向量机、AlexNET、GoogleNET、VGG-16NET效果最好,达到89.51%,实现对棉花的褐斑病、炭疽病、黄萎病、枯萎病、轮纹病、正常叶片的准确区分。该模型在棉花病害识别领域具备良好的分类性能,可实现自然条件下棉花病害的准确识别。

Cotton disease identification model based on improved VGG convolution neural network

In order to classify the cotton disease images accurately under natural conditions, a disease identification method based on the improved VGG-16 convolution neural network is proposed. Based on the VGG-16 network model, this method replaces the original three fully connected layers with two fully connected layers, and replaces the SoftMax classifier in the original VGG-16 network with the 6-tag SoftMax classifier. These changes can optimize the model structure and parameters, and it can share the weight parameters of the convolutional layer and the pooling layer in the pre-training model through the mini-migration learning. From the constructed image database of cotton diseases, the images of disease images were randomly taken as training set and test set to test the performance of the method. The experimental results show that this method can effectively extract multi-layer feature images of cotton leaf disease images, and it can highlight the edge information and texture information from cotton disease feature map through the Relu activation function. In the four kinds of resolution image samples training and verification, 512 pixels×512 pixels resolution images are the best fit for testing. Compared with BP neural network, support vector machine, AlexNET, GoogleNET and VGG-16NET, the proposed method has the best performance in average recognition accuracy of 89.51%. It is better to distinguish Cercospora leaf spot, Anthracnose, Verticillium wilt, Fusarium wilt, wheel disease and health leaf. The results show that the method has good classification performance in the field of cotton disease identification, and can realize accurate identification of cotton diseases under natural conditions.