Optimization Of Fruit Picking Order Based On U-Net Semantic Segmentation Model

In recent years,computer vision technology has been applied to agricultural production,and agricultural mechanization has been widely used.The application scenarios include fruit detection,fruit image segmentation,and fruit automation picking,among which the research on optimizing the fruit picking sequence is becoming more and more in-depth.The segmentation of fruit images is one of the important steps.When facing real-time picking scenes,Due to the unpredictable weather changes and the obstruction problem between fruits and tree branches,the accuracy of fruit image segmentation is greatly affected.The determination of fruit picking order is crucial for improving picking efficiency and reducing machine operating costs.It is necessary to have a certain detection speed to adapt to realtime picking scenarios.To solve the above problems,this article adopts an improved U-Net model to optimize image segmentation technology,By improving segmentation speed while maintaining good segmentation accuracy,a multi-objective grasshopper optimization algorithm based on clustering evolution mechanism was used to optimize the picking order of fruits.The specific work content is as follows:(1)A CA U-Net semantic segmentation network based on improved U-Net model is designed.This paper first compares the traditional image segmentation technology,introduces the current research status of image target detection technology and image segmentation,analyzes the characteristics and existing problems of real-time pear picking scene,and integrates the current target detection model and semantic segmentation model.U-Net,Deep Lab-V3 + and Mask R-CNN are compared in terms of segmentation effect,among which U-Net model has better segmentation accuracy and effect,but has a larger number of parameters.In this paper,the backbone feature extraction network VGG16 with a large number of parameters in U-Net is replaced by a lightweight Mobile Net-V3 network,which further improves the segmentation efficiency.The SE module in Mobile Net-V3 is improved to Coordinate Attention module,and the improved model improves the image segmentation accuracy under complex background,and has a good segmentation effect while maintaining a high segmentation speed.(2)A multi-objective grasshopper optimization algorithm based on clustering evolution mechanism is designed.In the process of determining the fruit picking order,due to the single condition of the traditional picking order determination method,the multi-objective grasshopper optimization algorithm is applied to the scene of fruit picking order optimization,and the fruit grouping mechanism is optimized by using the K-means clustering algorithm to group the fruits first,After that,the fruit picking sequence was optimized based on the two principles of the shortest picking distance and the smallest covered area of the fruit,and according to different picking scenes and different fruits,the weight of factors was reasonably allocated,and the improved swarm intelligence optimization algorithm was used to optimize the fruit picking sequence,and an excellent pear fruit picking strategy was designed.