Design of a Selective Transmission Framework in an Edge - Cloud Collaborative Object Detection Environment

This paper proposes a selective transmission-based edge–cloud object detection framework that combines a lightweight Single Shot MultiBox Detector with MobileNet backbone (SSD-MobileNet) detector on the edge with a high-precision Region-based Fully Convolutional Network (R-FCN) verifier in the cloud. The edge device performs an initial inference and computes a confidence score to control transmission: when the confidence exceeds a threshold, only the Region of Interest (ROI) is cropped and sent; otherwise, the full image is transmitted as a fallback to mitigate false negatives. By adapting the transmission unit (ROI vs. full frame), the proposed method reduces bandwidth consumption and cloud-side processing overhead compared to conventional full-image offloading. Experimental results demonstrate that the proposed framework achieves an Accuracy of 0.9815 and an F1-score of 0.9814, maintaining performance comparable to the cloud-only R-FCN model (Accuracy 0.9760). Furthermore, it reduces the average transmission data size from approximately 4.21 MB to 2.52 MB and achieves an end-to-end latency of 1.1178 s, which is superior to both the R-FCN standalone (1.2581 s) and SSD standalone models. These results confirm that the proposed method is suitable for real-time intelligent vision applications under resource- and network-constrained conditions.