Research On Detection Of Brassica Napus Growth State Based On Machine Vision

In agriculture,the growth state of plants affects the fruit and the harvest in the next few years.In order to master its growth state in time and control it artificially,it is necessary to detect its growth state quickly.The resistance of plant to diseases and insect pests can directly reflect the current and future growth state of plant.If the resistance is strong,the plant is not vulnerable to diseases and insect pests,and the growth state is good.If the susceptibility is strong,the plant is easy to be infected with diseases and insect pests,which has a great impact on the growth and development of plants.Therefore,this dissertation mainly combined with the knowledge and technology of machine vision to diagnose the resistance of Brassica napus to Pieris rapae,determined the resistance of Brassica napus to Pieris rapae,and then detected the growth status of Brassica napus.In order to classify the disease and insect resistance of Brassica napus and detect the growth status of Brassica napus,different image preprocessing,segmentation and feature extraction methods were studied in this dissertation.The classification table of disease and insect resistance of Brassica napus leaves,the classification table of disease and insect resistance of Brassica napus leaves and the diagnosis model of disease and insect resistance of Brassica napus leaves were designed.(1)Preprocessing of leaf image of Brassica napus L.The effect of image segmentation is mainly affected by image set and other factors,so it is necessary to pre process the image to improve the quality of image segmentation effect.In this dissertation,the method of color space model is used to segment the image of rape leaf target area.Compared with the common spatial models,the color components that can highlight the target area are selected for superposition,and then the thresholding segmentation is performed to obtain the complete plant leaf image.(2)Feature extraction and pest classification standard of leaf image of Brassica napus L.The feature extraction is mainly aimed at the shape features of leaves.According to the calculation and the standard of "rape germplasm resources description specification and data standard",the insect classification table was established to prepare for the follow-up work.(3)A diagnostic model for the resistance grading of Brassica napus leaves to Pieris rapae was established.Through the previous work,the 600 samples in this dissertation are graded,and the classified data are used as samples for training.The convolution neural network was used to design the classification model.The resistance category of Brassica napus leaves to be detected was set as the output,and the sample image was set as the input of the network to establish the convolution neural network model.By building vgg16 neural network model,the preprocessed image is input into the model,the cross entropy categorical cross entropy is selected as the loss function,the small batch gradient descent method is used to optimize the model,and the model is optimized through continuous repeated training.The actual simulation test results showed that the accuracy of resistance grading diagnosis for Brassica napus leaves was about 96.39%.The diagnosis model of resistance grading of Brassica napus leaves in this dissertation can quickly classify the resistance grades of Brassica napus and provide a new method and idea for the detection of plant growth status.