The prevalence of cheating in first-person shooter (FPS) games poses a formidable challenge, undermining user experience and the integrity of competitive play. In response to this issue, a visual-based anti-cheating detection network, termed VADNet, has been developed, harnessing the capabilities of deep learning and computer vision techniques. VADNet incorporates a focus module designed to segment high-resolution images, alongside a Feature Pyramid Network (FPN) for the fusion of multi-scale features, culminating in a classifier module tasked with the quantification of cheating behaviors. Rigorous experimentation on a dataset derived from a real online FPS game substantiates VADNet's efficacy in identifying players who resort to cheating, as evidenced by high precision, recall, and F1 scores. This investigation advances the field of anti-cheating mechanisms for FPS games, offering a robust and reliable system to preserve the fairness and integrity of online gaming environments.