A Multi-Information Detection Model for Violent Images based on YOLOv3-SPP and DenseNet
Ruifeng Guo, Wenyang Li, Hongliang Wang, Xiaoxing Zhang · 2022 11th International Conference of Information and Communication Technology (ICTech)) · 2022
The Internet has become an important carrier of people's daily life, and it is constantly developing and growing. In order to create a healthy and benign Internet environment, these sensitive pictures need to be identified and filtered. In the field of sensitive image detection, the violent image detection technology is still in an immature stage, and there is still a lot of room for development in the detection accuracy in the real world. Current research mainly uses classification models ResNet, DenseNet and target detection model YOLOv3 to detect violent images. However, the classification model has the problem of paying attention to the global information and ignoring the target's location information and local target information. Target detection has the problem of only identifying specific local targets and ignoring the global information. In order to solve this problem, this paper proposes a multi-information violence image detection model based on YOLOv3-SPP and DenseNet. Combining the information of the two models through the multi-information model improves the detection accuracy of violent images by 1.3% compared with DenseNet and 2.2% compared with the ResNet model.