Research On Leaf Disease Recognition Method Of Kiwi Based On Feature Fusion

Aiming at the problems of kiwi disease recognition difficulty and low recognition accuracy,this paper studies the disease recognition method combining traditional machine learning and deep CNN.The main research contents are as follows:(1)First,the kiwifruit diseased leaves to be identified are pretreated to ensure better identification results.Preprocessing mainly includes image noise processing and target area segmentation processing.In the noise reduction part of the leaf image,a local small matrix is used to average the pixel values to eliminate noise points.Convert the RGB color space image to the LAB color space image,then use the K-means segmentation algorithm to segment the diseased leaf background,and finally use the Otsu segmentation method for image segmentation to obtain the diseased spot image.Experiments show that the segmentation effect of this diseased spot segmentation method is compared it is good.(2)Feature extraction and selection.Feature extraction and selection are carried out for leaf images of three common diseases,and SVM classification model is used for disease recognition.MATLAB as a development tool,GUI interface design,a kiwi leaf disease recognition system is constructed.(3)Since kiwi fruit spots generally occupies a small area,it is difficult to use traditional methods to achieve accurate segmentation.Therefore,this paper proposes an improved VGG16 network to accurately identify them.In order to obtain a better training effect,directly use the trained model,and further adjust it based on its weight and bias.Using this method can greatly accelerate the training speed of the model.At the same time,the use of transfer learning methods for parameter acquisition can avoid the problem of insufficient leaf images to a certain extent.In the traditional VGG16 model,the selective convolution(Selective Kernel Convolution,SK convolution)module is introduced,and global average pooling is used to replace the fully connected layer,which solves the overfitting of the fully connected layer due to a large number of parameter optimizations.problem.Experimental results show that compared with other traditional network models,this algorithm can more accurately and quickly identify tiny spots on kiwi diseased leaves,which provides a certain theoretical basis for automatic control of kiwi disease..