A Feature Recognition System For Sugarcane Nodes Of Sugarcane Seed Cutter Based On Convolutional Neural Network

Sugarcane is an important tropical cash crop.Aiming at the development requirements of the sugarcane "double high" industry in Guangxi region,in order to promote the planting technology of sugarcane pre-cutting seeds with good seeds and good methods,combining with the technical index demand of the research group’s research and development of the intelligent transverse seed cutting machine for sugarcane pre-cutting seeds and the development trend of the current computer vision,in this paper,a sugarcane surface feature recognition system was constructed to identify and locate the characteristics of sugarcane surface stem nodes.The cutter cutting position was calculated according to the positioning information of sugarcane stem nodes,and the data of cutter cutting position was transmitted to PLC,implements the sugarcane pre-cut machine on the whole sugarcane internodes more continuous,dynamic intelligent identification and cut.The work of this paper mainly includes the following three aspects:(1)According to the growth of the collected sugarcane stem buds and the current development trend of deep learning,a convolutional neural network based sugarcane surface stem node feature identification and positioning model was proposed.The model is constructed by constructing feature pyramid structure,using data standardization,setting regression anchor and so on.After the test,the average AP value of the model for the identification of sugarcane stalk nodes was 90.4%,the average recall rate was 90.6%,the average precision rate was 96.9%,and the average identification time of a single sugarcane stalk node image was 28 ms.(2)The established sugarcane stem node identification and positioning model was improved,and its feature pyramid structure was reconstructed on the basis of the constructed convolutional neural network model.A new model was constructed by combining the shallow information with the deep information by using the combination of upper sampling and lower sampling.Add an additional loss of center distance to the prediction box in the original model’s loss function.After analysis and testing,the recognition rate of the modified model is 1.3% higher than the average AP value of the previous model.(3)According to the actual project requirements and technical indicators,calculated by the sugarcane stem section location information,cutting knife cutting location data,building of sugarcane cutting machine based on convolutional neural network recognition system realization of sugarcane stem section identification and calculating cutter cutting position transfer function,so that the model can process further data and transmit the data to the following seed cutting platform for real-time seed cutting.