Research On Key Technologies Of Rice Grain Image Recognition

Rice is the main food of the Chinese people,and food security occupies a pivotal position in national security.For a long time,the deterioration of food caused by weather and man-made factors has caused huge economic losses.At the same time,in order to effectively distinguish between high-quality rice grains and low-quality rice grains,a lot of money,time and energy have been spent on scientific research institutions and researchers in our country.In recent years,with the vigorous development of computer and network technology,many time-consuming and labor-intensive manual tasks in the past can be completed by machines.Therefore,in this thesis,a software for detecting the appearance quality of rice grains is studied,which can efficiently and accurately segment the rice grain images and classify the rice grains.The technology of rice grain image recognition is applied in practice.In this thesis,the following tasks are mainly completed:(1)Design a lightweight fast segmentation neural network LFSN(Lightweight Fast Segmentation Network).First,the watershed segmentation algorithm is used to segment the rice grain image.When the image segmentation algorithm based on traditional morphology cannot solve the tightly adhered rice grain image,a new type of neural network model LFSN is designed,which uses deep separable convolution,inverted residual structure,feature pyramid structure,and region suggestion network And other algorithms,it meets the multiple needs of rice grain image segmentation.(2)After realizing the image segmentation of rice grains,each grain of rice is classified into different categories such as perfect rice grains,moldy and diseased rice grains,yellow rice grains,and chalky rice grains.Mainly use neural network algorithms based on deep learning to compare different algorithms,and use model compression to increase the speed of the model to meet real-time requirements.(3)According to the research needs of the thesis,design and implement a rice grain appearance quality inspection software,which effectively integrates the rice grain image segmentation algorithm and classification algorithm,design and implement the main page,new task page,edit page,and effect display page.The main research of this thesis is to combine rice grain image segmentation and classification technology with software design,and effectively apply the research content of the thesis to actual engineering.