Research And Implementation Of Small Object Detection Algorithm Based On ResECNet

With the rapid development of deep learning technology,major breakthroughs have been made in the field of target detection,and target detection technology has been successfully applied in many fields such as security monitoring,intelligent driving,and medical image recognition.However,due to the characteristics of small targets(less pixels,easy to be confused by noise,etc.),the detection accuracy of current mainstream target detection algorithms for small targets is much lower than that of medium and large-sized targets.Aiming at the problem that small target features are difficult to extract,this paper improves the feature pyramid network and designs a global context information module to extract context information to assist in the detection of small targets.On this basis,the Res ECNet small target detection algorithm is proposed.The main research work of this paper is as follows:(1)Research on cross-resolution information interaction algorithm for small target detection.Aiming at the problem that the low-level features of the feature pyramid network are weak and the robustness is not strong enough to affect the detection of small objects,an improved form of the feature pyramid network is designed.On the basis of the original feature pyramid network,a feature splicing module and a bottomup path are added,and the features of each layer are further strengthened by fully fusing features of different resolutions.At the same time,in view of the problem that small target pixels are few and the extracted features are few,which makes it difficult to detect small targets,a global context information extraction module is designed to use global context information to assist small target detection.Finally,the global context information extraction module is fused with the improved feature pyramid network,so that the fused features contain both rich semantic and location information and rich global context information.(2)Small target detection based on Res ECNet.Aiming at the problems existing in the Retina Net algorithm,in order to better detect small targets,the Res ECNet small target detection algorithm is designed.This algorithm makes various improvements on the basis of Retina Net algorithm.It uses the deep residual network as the feature extraction network,uses the underlying feature map with higher resolution to detect small targets,and uses the fusion of global context information to re-optimize crossresolution information.The interactive network performs feature fusion and context information extraction,so that the final extracted features take into account deep semantic information and global context information;at the same time,an improved GIo U Loss is designed as the boundary regression loss of the algorithm,and K-means is used to set parameters for the anchor box to further Improve the performance of the algorithm for small target detection.Finally,the effectiveness of the Res ECNet algorithm is verified by experiments.(3)Design and implementation of small target detection prototype system.In order to actually deploy the Res ECNet algorithm designed in this paper,this paper designs and implements a small target detection prototype system based on C/S architecture.The prototype system uses Res ECNet as the small target detection algorithm,which has user login,image detection,video detection,data storage,detection Features for record viewing,model training,and user rights management.Finally,the system is tested to verify the effectiveness of the system.