Study On Sugarcane Stalk Joint Recognition And Cutting Based On Convolutional Neural Network

Sugarcane is an important sugar crop and production fuel.The automation of sugarcane pre-cut seeds is a key part of realizing mechanized sugarcane planting.Sugarcane stem node recognition is an important research for sugarcane seed cutting.Only by identifying and positioning the sugarcane stem nodes can the next step of cutting the stem nodes be carried out,so as to realize the automation of sugarcane seed retention.This can not only reduce the labor intensity of sugarcane planting,but also reduce the physical damage to the stem nodes during the process of cutting seeds.In this paper,a sugarcane seed cutting device that automatically recognizes sugarcane knots is designed,and the improved YOLOv4 sugarcane stem node recognition method is constructed to identify the characteristics of the surface of the sugarcane.According to the information of the stem node,the seed cutting device transmits the sugarcane,and Transmit the cut-to-position data to the PLC to realize stable and continuous seed cut.The main research contents are as follows:(1)Designed and developed a sugarcane seed cutting device that automatically recognizes sugarcane stem nodes.Through image acquisition technology and a stable sugarcane conveying system,the sugarcane can be transported stably without damaging the stem nodes,and the collected information can be sent in real time.To the control system,the control device continuously cuts the sugarcane stem nodes in real time.(2)Collect black-skinned sugarcane and green-skinned sugarcane stem node samples,including special conditions such as darker environment,natural light environment,mud and part of the stem node damage,etc.,and expand the sample and construct a sugarcane stem node data set.(3)Recognize and test the sugarcane stem node data set based on YOLOv4.In order to further improve the accuracy and speed of model recognition,a sugarcane stem node recognition model based on improved YOLOv4 is proposed.The effective feature layer obtained through the backbone feature network is directly passed into the enhanced feature extraction network for path aggregation construction;the Mish activation function in the original model is changed to the Leaky Re LU activation function,the accuracy of the improved model does not decrease,and the recognition speed is 2 times the original.(4)Perform static and dynamic recognition on selected sugarcane stem node samples.Compare the recognition accuracy and recognition speed of the original YOLOv4 and the improved YOLOv4 model.The improved YOLOv4 model has a higher recognition accuracy and the recognition time is the second of the original model.One part.It is verified through experiments that the seed cutting device can better complete the transportation,identification and continuous seed cutting operations of sugarcane.