Small target detection method based on feature fusion for deep learning in state grid environment evaluation

Abstract

Aiming at the problem that small and medium-sized targets cannot be detected in real time in high-resolution images, a new target detection network model is proposed. Firstly, the residual network RESNET is used as the basic network structure, an additional pyramid network model is added, and the pool layer is used to increase the number of hierarchical feature mapping. Then, the feature map is deconvoluted, and the high-level semantic feature map information and shallow feature map information are fused. Finally, the target is detected. Based on the analysis of the experimental results, compared with the existing target detection network model, the deep learning network model using feature fusion techniques has a detection accuracy of 80.2% on the standard dataset Pascal voc2007, and the detection speed reaches 27 frames per second, which meet the requirements of high-resolution image real-time monitoring and small target detection.