The demand for multi-target detection within an IoT-based edge-cloud surveillance system is increasing. This is particularly the case in real-world scenarios where there could be several targets in varied lighting and several very mobile objects. Even with the best possible models, object detection models collapse when presented with the randomness of real-world environments, including clutter and the detection of multiple objects within a scene. A new innovation, the Enhanced Hyper-node Faster Relational YOLO Dwarf Mongoose (IHnode-FRYDM) Graph Attention Network (GAN) for multi-target detection in IoT-based innovative edge-cloud surveillance systems is presented herein. The new method uses the PASCAL VOC dataset to create a more efficient detection framework. It starts with the Iterative Dependable Peak-Aware Directed Filtering (IDPADF), a newer technique for pre-processing images, that considerably improves both the input image and feature representation quality. The real detection then executes the Faster-YOLO architecture, which is essential since it strives to balance speed and accuracy for real-time IoT operations. Moreover, it uses a Hyper-node Relational Graph Attention Network (HRGAT) to perform effective relational feature learning and correct identification of multiple targets in intricate and dynamic environments. IDMO's performance maximizes the rate of convergence and stability of the model to meet the computational loads of IoT edge devices. The resultant evaluation provides a mAP of 99.6% and an F1-score of 99.5%, while offering a processing time reduction of 32% in comparison to other traditional approaches. The results suggest that the new framework can be successfully deployed into new IoT edge-cloud surveillance processes with an efficient and accurate process to fulfill technical demands of multi-target surveillance applications.
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号