Research On Garbage Detection Algorithm Of Surface Cleaning Robot

With the improvement of people's living standards,the amount of garbage produced by people's life and production is also increasing.Because some people's environmental awareness is not strong,some of garbage is dumped in the water,if not cleaned up in time,it will have a serious negative impact on people's normal life and production.The traditional manual cleaning method of surface garbage is not only time-consuming and labor-intensive,but also extremely dangerous.A intelligent surface garbage salvage robot emerges at the right moment,and the detection of surface garbage is the focus of the research on intelligent surface cleaning robot.This paper focuses on the research of the detection algorithm of the surface garbage,and applies the computer vision technology to the detection of the surface garbage in order to realize the automatic detection of the surface garbage.This paper introduces the latest progress of the research on the detection algorithm of the surface garbage.Aiming at the non-surface area and the interference factors of the surface contained in the image acquired by the cleaning robot in the operation,the detection of the surface garbage is divided into three stages: surface segmentation,garbage object detection and segmentation detection fusion.The first stage is surface segmentation,a surface segmentation network composed of a lightweight neural network based on Mobile Net V3-Large(1.0)and an improved semantic feature fusion module of Atrous Spatial Pyramid Pooling(ASPP)is used to segment the images acquired by the cleaning robot into surface and non-surface regions;The second stage is garbage object detection,which is based on the standard version of You Only Look Once-v3(YOLOv3),and the Generation Intersection over Union(GIo U)and the improved feature fusion layer are used to improve the accuracy of the garbage detection network.At the same time,network slimming algorithm is used to improve the detection speed of the network;the third stage is segmentation detection fusion,which combines the surface segmented from the image in the first stage with the garbage object detected in the second stage,so as to eliminate the garbage in the non-surface area and finally realize the surface garbage detection algorithm.Through experimental verification and testing,the average Pixel Accuracy of the proposed surface segmentation network on the testset of surface segmentation is 95.9%,and the Mean Intersection over Union can reach 94% and segmentation speed can reach 56 FPS.The average accuracy of the improved YOLOv3 garbage detection network is 4.8% higher than that of the standard version of YOLOv3 on the testset of garbage detection,and the detection speed is nearly doubled.The detection accuracy of the whole surface garbage detection algorithm after the fusion of segmentation detection results can reach 77% on the data set only marking the surface waste object,and the speed of detection can reach 32 FPS.The experiment and test results shown that the algorithm proposed in this paper can detect the garbage target on the surface of complex environment,and has certain engineering application value.