Research On Vision-based Autonomous Identification System For Crop Diseases And Insect Pests

In order to better identify the image information of crop diseases and insect pests,taking corn spot as an example,an unmanned airborne machine vision system is used to collect corn spot in the open air.Based on the collected images,anisotropic denoising enhancement and maximum The method of entropy segmentation is compared with the method of convolutional neural network.For the methods of anisotropic denoising enhancement and maximum entropy segmentation,the cropped image was pre-processed with histogram equalization and gray-level enhancement,and the factors affecting the segmentation of the lesion target were analyzed,and then anisotropic diffusion denoising Method,denoising and enhancing corn spot image.The maximum entropy segmentation method was used to achieve the feature segmentation of the corn lesion image,and the morphological analysis was used to accurately extract the lesion image.Experiments on convolutional neural networks using a self-made training set of convolutional neural networks.Based on Alex Net,a suitable convolutional neural network was designed.The dropout layer was introduced to solve the gradient dispersion problem when the forward data stream was propagated,and the Adam optimizer was introduced to optimize the loss function when the reverse loss function was propagated.Through comparison,analyze their advantages and disadvantages and improvement methods.By analyzing the process and results of the two methods,it is not difficult to find that deep learning can skip the complex feature extraction step when identifying crop diseases and insect pests,but there are higher requirements on the collection of data sets,and because of their hidden The characteristics of the type recognition,its characteristics can not be accurately displayed;the image segmentation processing method can directly process and analyze the image to be recognized,and clearly show the recognition process,but the calculation process of feature extraction More complicated.The experimental results prove that the deep learning method can also be used to quickly and easily recognize crop disease and pest images.Figure 25;Table 3;Reference 59...