| With the development of information technology,electronic evidence based on computer and network plays an increasingly important role in proving the facts of cases.In the past,the identification of electronic evidence mainly relied on manual labor,which was not only time-consuming and labor-intensive,but also prone to errors and difficult to adapt to new situation.In order to solve these problems,some scholars have proposed to apply deep learning technology to the field of image electronic evidence recognition and screening.Target detection and recognition algorithms based on deep learning can be roughly divided into two different types.One is a two-stage algorithm for stepwise regression and classification of candidate frames,such as typical R-CNN,Faster RCNN;the other is a single-stage algorithm that does not require target regression and classification generated by candidate frames,such as typical SSD,YOLO.Using drug-related image electronic evidence identification as the application background.Deep researches have been conducted on image recognition algorithms based on Faster RCNN and SSD,and the main research contents and innovations are as follows:(1)Aiming at the shortcomings and problems of the classic two-stage target recognition algorithm Faster RCNN in drug image target detection and recognition,an improved Faster RCNN drug image target detection and recognition algorithm is proposed.Specifically,a feature pyramid is constructed for multi-scale feature fusion and feature extraction,non-maximum extremum suppression Soft-NMS is introduced,and ROI Align technology is used to avoid accuracy loss caused by multiple quantization.Through the detection and recognition of targets in 8 types of different drug-related images,the results show that compared with the original Faster RCNN model,the improved model has a certain improvement in the accuracy of drug image target detection.(2)Aiming at the shortcomings and problems of the classic single-stage target recognition algorithm SSD in drug image target detection and recognition,an improved SSD drug image target detection and recognition algorithm is proposed to improve the shallow layer of the algorithm in the SSD network structure.The convolutional block attention module is added after the convolution layer to further improve the characterization ability of the feature map;the image features obtained by the convolution of different convolution layers are multi-scale fusion,and the shallow feature information is added.Using this algorithm to detect and recognize targets in 8 types of different drug-related images,the results show that the average accuracy of the improved SSD algorithm is increased by 4.7%.(3)Based on Tensorflow deep learning framework,a drug image electronic evidence recognition system is designed and implemented,including four functional modules: system management,system monitoring,model training management and drug image recognition management.GPU is used to accelerate the model training process,improve the speed of model training and testing and the stability of image recognition system. |
