Research On The Detection System Of Cucumber Leaf Diseases Based On Machine Vision

Plant disease is one of the main factors that harm the quality of vegetables,which damages farmers' economic income to a great extent.Cucumber is easily threatened by diseases during its growth cycle.Large-scale and high-density planting methods bring great challenges to traditional disease control measures.The traditional methods of disease prevention and control are through human eye observation or infection experiments.This method has the disadvantages of low efficiency and strong subjectivity,and excessive use of chemical pesticides will not only lead to pesticide residues,plant resistance problems,but also environmental pollution.Machine vision,artificial intelligence and other related technologies have been widely used in agriculture,especially in plant disease identification and fruit quality detection.In this paper,cucumber leaf disease images with different disease degrees are taken as the research object,and the research on the classification and identification method of cucumber leaf disease degrees by using machine vision,deep learning and other relevant technologies is of great significance to improve the quality of agricultural products and ensure the green and sustainable development of agricultural ecological environment.In this paper,the classification and recognition methods of different disease degrees of three main cucumber leaf diseases in the natural background were studied,which mainly included the following contents:(1)The images of cucumber leaves with different degrees of disease were processed by data enhancement,through rotation,distortion,image transformation,adding noise and other geometric transformation methods without changing the categories and properties of the images.The data set samples were expanded to solve the problem of insufficient training samples,and the over-fitting problem of the model was reduced in the training,and the created data set was divided into training set and test set according to a certain proportion.(2)In order to avoid the complex image preprocessing and artificial feature design process in traditional recognition methods,a Convolutional Neural network model is constructed based on the vgg-16 network model.In this algorithm,micro-migration learning and parametric rectification function(PReLU)are firstly used to construct deep convolutional layer,which can quickly and steadily extract the features of cucumber leaf disease input images of different sizes.Then the output representation of cucumber disease feature image with fixed size was realized by using the method of spatial pyramid pooling.Finally,the parameters in the full connection layer were replaced to reduce the computational complexity,and the degree of cucumber leaf disease was identified by Softmax.Experimental results show that the network model proposed in this paper has good recognition performance,and the classification recognition rate of cucumber leaf disease degree is 87.24%.(3)Aiming at the over-fitting problem of CNN in the training process,the influence of the size of convolution kernel,the optimization algorithm of gradient descent and the parameter selection of activation function on the over-fitting rate of the model was analyzed and discussed,and the parameters in the network model were adjusted for experimental comparison to verify the image classification and recognition performance of the improved deep CNN model.