Research And Design Of Garbage Classification Detection System Based On Convolutional Neural Networ

With the rapid development of economy and society,people ’s awareness of waste resource recycling and reuse has been strengthened.Garbage classification and recycling is a key link in waste recycling and reuse.Most of the existing garbage classification tasks rely on manual classification of garbage.Research on using deep learning technology as a garbage classification scheme is not common enough.This is mainly because the morphological differences of garbage targets are large,feature extraction is difficult,and garbage classification has a wide range of application scenarios,which requires high computational power for detection equipment.In order to solve the above problems,this paper applies deep learning technology to the task of garbage classification and detection,classifies and locates the garbage in the image through convolutional neural network,and designs and develops a garbage classification and detection system to verify the performance of the model.The results show that the system can meet the requirements of practical application.The main research contents are as follows:(1)Research and analyze the public data sets in the field of garbage classification and collect the garbage images that meet the standards in the public data sets.At the same time,the data sets are expanded by means of manual shooting and network search to obtain single-target and multi-target garbage image data sets for classification and detection tasks.(2)A lightweight garbage classification network is proposed.Mobile Net V3 is used as the baseline network of the classification task and the network structure is improved.The standard static convolution in the original Mobile Net V3 network is replaced by full-dimensional dynamic convolution,and multi-dimensional feature information is extracted.The training is performed on the created single-target garbage classification data set,and the Res Net18,VGG16 and Mobile Net V3 network models are used for comparison experiments.The experimental results show that the model has high accuracy and good classification effect.(3)An improved target detection network structure based on FCOS algorithm is proposed.Based on the FCOS algorithm,the pyramid attention upsampling module is used to improve the feature extraction ability of the network.The adaptive sample selection method is used to improve the allocation strategy of training samples.The original IOU Loss is replaced by GIOU Loss to improve the accuracy and convergence speed of the model.Experiments are carried out on the self-made multi-target garbage detection data set.The results show that the algorithm has improved detection accuracy and detection speed.(4)A garbage classification and detection system is designed and completed,and two modules of classification and detection are realized.Through the verification experiment of the system,it is proved that the system can meet the requirements of practical application.