An Improve MobileNetV2 For Monitor Image Anomaly Detection
Xinlin Yang, E. C. Lee, Jiaqi Huang, Qin Guan · 2023
In the field of video surveillance, due to signal interference, poor environmental conditions, data transmission and other reasons, so that the surveyed video contains a variety of noise. If these noise types cannot be effectively identified, targeted denoising cannot be carried out, thus reducing the accuracy of subsequent recognition. To solve this problem, this paper proposes an improved MobileNetV2 network structure, which uses SE module to reconstruct the original block. We propose three different reconstruction methods, using these three reconstructed blocks to build MobileNetV2 can make the network have stronger feature learning ability. It can effectively detect a variety of noise types, and the detection accuracy is better than MobileNetV2.