Research On Target Detection Based On Multi-scale Characteristic Map Fusion

In natural scene images,the quality of feature extraction is the key factor to determine the performance of target detection.At present,most detection algorithms use the powerful learning ability of convolutional neural network to obtain the prior knowledge of the target and carry out the target detection according to the knowledge.The low level features of convolutional neural network are characterized by high resolution ratio,low abstract semantics,little position information and lack of representation of features;High-level features are characterized by high identification and low resolution ratio,and the ability to detect small-scale targets is weak.Based on convolutional neural network,this paper proposes a multi-scale feature graph fusion target detection method.Firstly,the original SSD(Single Shot MultiBox Detector)network is used to extract the feature graph.The extracted feature graph is unified into 256 channels through 1?1 convolution layer.Secondly,the spatial resolution of the top-down feature maps is increased by deconvolution.Then,the feature graph in two directions is fused by adding corresponding elements.The fused feature graph is convolved with the convolution kernel of 3?3 to reduce the aliasing effect of the fused feature graph.According to the above steps,the feature graph with strong semantic information was constructed and the details of the original feature graph were retained.Finally,the aggregation target is selected through the prediction box,and the final detection is achieved by non-maximum Suppresion(NMS,Non-Maximum Suppresion)algorithm.The experimental test results of the PASCAL VOC 2007 and PASCAL VOC 2012 datasets show that the mAP(mean Average Precision)of this method is 78.9% and 76.7%,which is 1.4% and 0.9% higher than the classical SSD(Single Shot MultiBox Detector)network;In addition,when detecting small-scale targets,the detection effect is improved by 9.3% compared with the classical SSD algorithm,achieving the goal of accurate target detection.