Digital Forensics-Based Consumer AAV Anti-Attack Detection Algorithm Using Time-Frequency Feature Fusion for Pine Wilt Disease

Hong Ye, Xiajun Ding, Nazik Alturki, Ali Kashif Bashir, Meilei Lv, Kai Fang, Wei Wang

Abstract: Early detection of Pine Wilt Disease (PWD) with consumer Autonomous Aerial Vehicle (AAV) is vital for preventing its spread. However, malicious attacks on AAV recognition models can result in missed detections and false alarms, undermining disease control efforts. To address this challenge, we propose YOLOv8-FB, a AAV anti-attack detection algorithm for PWD. By integrating digital forensics methodologies, our approach mitigates adversarial interference while enhancing traceability. These advanced anti-attack strategies also show promise for biometric systems and consumer AAV data analysis, where robust detection under adversarial conditions is paramount. First, we propose a Frequency Domain Downsampling (FDD) module, which performs three types of pooling operations on the input feature map during downsampling. Average pooling and max pooling smooth image features, while wavelet pooling reduces high-frequency noise caused by adversarial attacks, thereby improving the model’s anti-attack capability. Second, we incorporate a Bi-Level Routing Attention (BRA) module to enhance the model’s perception of small targets. The BRA module, through dynamic sparsity and bi-level routing mechanisms, enables the model to adaptively focus on target regions and suppress the interference of background noise. When evaluated under a MIM attack with an intensity of 0.8, our model achieved a average precision improvement of 8.4%, 6.8%, 7.1%, 14.2%, and 8.2% compared to five most representative models.

External IDs:dblp:journals/tce/YeDABLFW25