Design Of Ground Garbage Detection Algorithm Based On Image Processing

Plastic bottles,cans,paper towels,cigarette butts and so on are common garbage in the streets.In order to ensure the cleanliness of the road and the good mental outlook of the city,many environmental sanitation workers need to inspect and pick up ground garbage.At the same time,as living standards are getting better and better,medical equipment is becoming more and more perfect,human life expectancy continues to increase,and the cost of employing units also increases.Therefore,in order to solve the above problems,it is necessary to design a road garbage detection system that can replace sanitation workers to identify and locate garbage.The idea of this paper is to use deep learning theory to comprehensively realize the identification and positioning of garbage,and then use small robots to grab ground garbage.The main research work includes :(1)Overall scheme design of ground garbage detection systemFirstly,according to the actual industrial production and the requirements of the whole system detection accuracy,accuracy,stability and low cost,the overall scheme of the ground garbage detection system is designed.Secondly,the depth camera,auxiliary light source and lighting method in the image acquisition module are selected and calculated.Finally,the small robot of the execution module in the whole system is selected,and the end fixture is optimized.(2)Spatial location of ground garbage imageThe parameters of RGB camera and depth camera are obtained by calibrating the depth camera.At the same time,the conversion relationship between RGB camera coordinate system and depth camera coordinate system is derived theoretically,so that the alignment between the depth map and RGB image can be achieved;Finally,the depth measurement accuracy of the camera is experimentally verified through the Depth Quality Tool kit.(3)Ground garbage enhancement algorithm designDue to the influence of ground environment,light,rain and other factors,combined with the actual scene of ground garbage detection,this paper analyzes the low illumination pictures and unclear pictures that may appear in the process of camera shooting,and improves on the basis of the original Retinex algorithm.The illumination component and reflection component are adjusted by the homomorphic filtering algorithm to enhance the image brightness and restore the real image information.The hue component,saturation component and brightness component are obtained by HSV color space conversion and processed separately for each part.By histogram equalization algorithm,gamma correction and other enhanced image contrast,enhanced image texture information.(4)Target detection algorithm design on ground garbageIn order to detect ground garbage,this paper proposes an improved Faster RCNN(Faster Region-based Convolutional Neural Network)ground waste detection algorithm.In the backbone network,resnet50 is selected as the feature extraction network,and the Squeeze-and-Excitation Networks(SENet)is added to strengthen the extraction of useful information in different channels.At the same time,the shallow information in the network is extracted pertinently to improve the detection accuracy of small targets in ground garbage.Finally,the feature pyramid network(FPN)is used to strengthen the feature fusion between multiple outputs of the feature extraction network;In the part of regional proposals network(RPN),through the statistical analysis of the data such as the scale and width height ratio of the data set,an anchor generation mechanism more suitable for the ground garbage data set is proposed;In the part of detection and prediction,the ROI align structure is used to replace the ROI pooling in the original paper to reduce the error in the process of feature map mapping and averaging.At the same time,through the "cascade" structure,the threshold of IOU is increased in two stages,so as to improve the accuracy of candidate boxes.The experimental results show that the model m AP is 96.6% by using the proposed algorithm,which is increased by9.6%.At the same time,the missed detection and false detection are reduced.(5)Ground garbage attitude detection algorithm designBased on the deep learning method,the algorithm framework of attitude detection in this paper is designed.The attitude direction angle is output by 18 classification labels,and the 4-dimensional grasping matrix(x,y,w,h)is output by regression method to improve the accuracy and speed of attitude detection.At the same time,the Inception V2 network is improved by using multiple small convolution kernels instead of large convolution kernels to improve the detection speed.After experimental testing,the detection accuracy of this algorithm can reach 90.72%,and the detection speed can reach 18.72 fps.