Research On Sugarcane Stem Segment Recognition And Seed Cutting System Under Transverse Transport Based On Computer Vision

Sugarcane is an important raw material for sucrose production in China,but the whole sugarcane industry has a low degree of mechanization in the planting link,which leads to low efficiency,high cost and lack of competitiveness in the international market.At present,the main planting method of sugarcane is pre-cut seed planting.Compared with real-time cut seed planting,it has lower bud injury rate,higher planting efficiency and less labor requirement.However,at present,pre-cut seed planting mainly relies on manual pre-cut seed,which leads to low overall efficiency.Therefore,it is urgent to develop an intelligent seed identification and cutting system for bud prevention.Aiming at the problems that the current sugarcane stem segment identification and seed cutting system cannot identify the whole sugarcane stem segment at one time,and the poor working condition of agricultural machinery leads to the decrease of accuracy after the interference of recognition,this paper proposed a sugarcane stem segment identification and seed cutting system based on computer vision under transverse transportation.Specific research contents are as follows:(1)Overall scheme design of sugarcane stem section identification and seed cutting system under transverse transportation.Aiming at the problem that the field of vision of ordinary industrial cameras is limited and the overall size of sugarcane is long when conveying along the transverse,the whole sugarcane global image that meets the recognition requirements cannot be captured,a visual acquisition module using two cameras is designed,and the hardware selection and construction are completed.According to the requirement of identifying and cutting seed system,complete the overall process scheme design.Finally,Zhang Zhengyou calibration method was used to complete the camera calibration.(2)Research on sugarcane image stitching method under transverse transport with improved SURF algorithm.Aiming at the problems of smooth sugarcane surface and low matching accuracy of extracted feature points,artificial marker elements were introduced into the background to assist registration,so as to improve the registration accuracy during stitching.To solve the problems of uneven distribution of extracted feature points and slow stitching speed caused by extracting global feature points,a mesh segmentation feature extraction method based on SURF algorithm was proposed.In order to improve the accuracy of image Mosaic,a new image registration method based on double screening is proposed.Finally,the image fusion method based on the optimal suture line is introduced to solve the problem of double image caused by stitching to some extent.(3)Research on sugarcane stem node identification under the complex background of improved YOLOv5.Aiming at the low accuracy of stalk recognition caused by background clutter in the recognition process,a cane stalk recognition method based on improved YOLOv5 was proposed.The backbone network and the neck feature map of the same size are weighted across layers,and the weights of different input feature layers are constantly adjusted in the iterative process to enhance the information fusion between different levels.The model loss function is improved.On the one hand,EIo U loss function is introduced to replace the original CIo U loss function to improve the accuracy of boundary frame regression.On the other hand,Focal Loss function was used to replace cross entropy loss function to solve the problem of unbalanced ratio of positive and negative samples.Finally,Ghost module is introduced for lightweight design of the model.The experimental results showed that compared with the original model,the average accuracy of the model proposed in this study was increased by 1.4%to 97.8%,the single piece detection time was 16.9ms,and the model size was only 11.40 Mb,which realized the identification of sugarcane stem nodes under different clutter levels and reduced the influence of background clutter during seed cutting.(4)Complete the human-computer interaction interface design based on Py Qt framework,and carry out stem joint identification experiments on sugarcane with different varieties and different complexity backgrounds.The data processing algorithm of cutting position was completed,and the cutting position was transmitted to the cutting platform for cutting seed experiment.The final qualified rate of cutting seed was 98.31%,and the throughput of stem node was 6277 bud segments/hour.