Method And Software Development Of Cucumber Downy Mildew Image Recognition Based On Deep Learning

Image recognition technology belongs to a pattern recognition technology and is an important field of artificial intelligence.It refers to the recognition of images.It is a technology used to identify objects and goals of various patterns.It can simulate the process of visual cognition and understanding through modern information processing technology and computer technology.Its main content is to extract the characteristics of the image,based on image processing,and based on this image recognition and classification.In this paper,we present a new method for image recognition of cucumber frosting mildew based on deep learning neural network:1.The experiment observed the changes of disease spots on the leaves by adjusting different environmental conditions such as different varieties,different temperature and humidity,different leaf ages,and whether the leaves are moist or not.The difference between image samples of frosting mildew under various conditions was found.Different image samples are provided for software recognition.The experimental results show that: The difference of resistance and leaf age has a great influence on the speckle image of downy mildew.The susceptible species and the strong age leaves are the most serious after the disease,and the mildew layer on the back of the leaves is dense.The spot shows a polygonal state,and the spot area is large.And concentrated;Under the condition of 25 °C and 95 % humidity in different temperature and humidity tests,the leaves have the longest survival time and the most continuous spot changes.In the experiment of leaf wet or not on leaf spot change,it was found that the leaf with wet surface had a slightly faster rate of disease spot than the leaf without wet surface,but the overall difference was not significant.2.A large number of healthy leaves and leaves infected with frost mildew were collected for sample processing.200 samples were randomly taken as test samples,300 as test samples,and the remaining 3,874 images were used as training samples.In this study,the samples of cucumber leaf diseases were pretreated,4,374 samples were obtained,and an 8-layer neural network model was designed.In sample training,set the accuracy of the model and the size of the picture during training,the number of pictures per batch,the number of steps per round of training,the number of rounds of training,and the test of the number of rounds verified.The most important effect on the preparation rate was the size of the input image.It was also found that in the picture size 256×256,the number of pictures in each batch was 128,the number of steps per round was 50,and the number of iterations was 15.The highest accuracy rate is 50 times,and the final result accuracy rate reaches 94.5 %,which can be used for the identification of actual situations.