Research On Multi-target Detection Of Dog Face Based On Improved SSD Algorithm

With the development of society,people's living standards are also improving,and dogs appear more and more frequently in people's daily life.More and more institutions and organizations have invested in the research and application of dogs.Among them,the multi-target detection of dog's face has very important research significance and application scenarios,which can be used for the management of breeding dogs in kennels,the management of stray dogs in cities,and the pet dog feeding in individual families.At present,deep learning is the useful algorithm in the field of object detection.Convolutional neural network(CNN)has achieved good results in face detection of dog.However,the current object detection algorithms in the detection of small targets and occluded targets are not satisfactory.In addition,it is very difficult to complete the detection of dog's face when dog's face in complex background.In the engineering application of dog's face detection with the requirement of timeliness,the detection algorithm must have the ability to detect small target and occluded target,the ability to detect target in complex background and the ability to detect target quickly.In order to meet the requirements of dog's face detection engineering,this paper is based on SSD algorithm and improves SSD accordingly.On this basis,the innovative work of this paper mainly includes the following aspects:(1)Aiming at the problem that the ability of SSD network to detect occluded target and small target of dog face is poor,this paper proposes a DC-SSD network based on pyramid dilated convolution.The DC-SSD network improves the shallow feature layer and the middle deep feature layer of SSD,and adds the pyramid dilated convolution and the feature pyramid dilated convolution,which effectively improves the detection ability of occluded targets and small targets.(2)Aiming at the problem that the ability of SSD network to detect targets in the complex background is poor,this paper proposes a DC-Attention-SSD network based on attention mechanism and feature fusion.DC-Attention-SSD network filters the features of dog face,reduces the influence of background features from space and channel,and uses featurefusion to enhance the information of dog face features,which effectively improves the accuracy of target detection in complex background.(3)In order to improve the detection speed of dog face detection engineering,this paper proposes to use lightweight convolution instead of traditional convolution to improve the detection speed.Compared with the traditional convolution,the special structure of lightweight convolution can guarantee the performance of the network,effectively reduce the amount of network operation and improve the detection speed.