Research Of Sugarcane Identificationbased On Computer Vision

Sugarcane node identification is the key technology for sugarcane planting mechanizing.The automation of sugarcane plant cutting can be realized,and then the sugarcane planting process can be automated only if the node of sugarcane has been recognized efficiently.There are some problems for the sugarcane identification: such as the complicated background,different colors of various kinds of sugarcanes,the chaos texture on the skin of the sugarcane.To address these problems,this paper proposes three solutions,which are based on image process,structured learning and deep learning respectively.The research contents and conclusions are as follows:(1)The sugarcane image dataset constructing.Two kinds of the sugarcane(black,yellow),collected in both day and night time,were photographed under different backgrounds.The sugarcane image were rotated by ±10?±20?±30 degrees.A total of 800 images were randomly selected from the final image dataset,and labeled manually.(2)Target region extraction.There is coarse skin,chaotic texture in the sugarcane target region and the sugarcane presents approximately linear tendency in the image.An iterative linear fitting algorithm was designed to extract the sugarcane target.(3)The study on sugarcane node recognition.Three sugarcane node identification solutions were proposed,which based on wavelet transform technology,structured leaning,and deep learning respectively.In the first solution,we used wavelet technology to decompose the image,and then spent much time to analyze various of reconstruction strategies.Then an optimal strategy was picked for node identification.In the second solution,a structured random forest was trained.Based the edge probability image generated by the random forest,an heuristic edge algorithm was used to detect the edges near sugarcane node.In the third solution,a convolutional neural network was designed and then trained.The heuristic edge algorithm was used again to detect the edges near sugarcane node.Four hundred of sugarcane was used in the experiments.The experiment verifies the difference of sugarcane image edge between the sugarcane node and sugarcane section is useful for the sugarcane node recognition.and the experiment al results shown that the recognition rates of three solutions were 92%,93%,94%,and the time consumption on one image is 1.96 s,1.50 s,0.90 s respectively.Comparatively the 3th solution spend less execution time.