Achieving effective multi-target detection in edge-cloud environments within the Smart Internet of Things (SIoT) poses a number of issues, especially in dealing with dynamic IoT sensor data, real-time data processing, and suboptimal parameter optimization. The existing methods often fall short due to poor optimization, leading to lower detection accuracy and increased computational overhead. These issues can be categorized into two primary problems: (i) suboptimal optimization of parameters, which affects detection precision, and (ii) inadequate real-time processing of large-scale and dynamic IoT data, which leads to delays and inefficiencies. To address these challenges, this paper proposes an Optimized Auxiliary Physics Informed Neural Network for Multi-Target Detection in Edge-Cloud within Smart Internet of Things (APINN-MTD-EC-SIoT). Here, the data is collected from the PASCAL Visual Object Classes (PASCAL VOC) dataset. The initial phase involves performing multi-target detection at the edge layer by analyzing sensor data from IoT devices. A Sparse Reconstructive Multi-View Evidential Clustering (SRMVEC) segmentation method is employed to refine the object detection, especially for moving objects in surveillance. Following this, the features such as edge, corner, texture, and color histogram are extracted using Singular Value Decomposition-Based Graph Fourier Transform (SVD-GFT). Then, the APINN classifies objects, including bags, bottles, and phones with less computational time. The APINN weight parameters are optimized by the Humboldt Squid Optimization Algorithm (HSOA) for enhancing accuracy. The performance metrics, like accuracy, precision, recall, and F1-score is evaluated. The APINN-MTD-EC-SIoT achieves higher accuracy of 99.96% and improvements in these metrics when compared to the existing methods.
