Efficient Tensor Compression and Reconstruction in Split DNNs for Edge-Based Object Detection

Computer Vision (CV) tasks are among the most pivotal, yet challenging, operations for uncrewed aerial vehiclesm (UAVs), especially in mission-critical applications. They require processing complex image data through Deep Neural Networks (DNNs), which demand computational resources far beyondUAVs’ capacity. To address this limitation, Split DNNs offer a promising solution by partitioning the model into: (i) a lightweight Head, deployed on the UAV for rapid, albeit less precise, initial image representations, and (ii) a more complex Tail, executed at the network edge for refined, higher-accuracy results. However, this solution necessitates transmitting large sensor data from the UAV to the edge server, leading to significant bandwidth consumption. We tackle this challenge by introducing a goal-oriented framework named Compressed Tensor-based DNN Split (CoTeD). Our framework integrates an applicationand system-aware optimization model that orchestrates computing and transmission resources in real time. At the UAV, CoTeD dynamically selects relevant tensor information and optimally compresses it, guided by application requirements and system operational conditions. At the edge server, CoTeD reconstructs the tensor, enabling efficient inference by the Tail model. This approach effectively balances bandwidth usage with quality of the CV task output. Experimental results, obtained throughour hardware-software testbed and using datasets with different sizes and characteristics, show that CoTeD can reduce data transmission over the radio link by up to 90% without notable loss in object detection quality and reduces inference latency by up to 70% compared to local DNN deployment on the UAV. Additionally, compared to static JPEG compressions, CoTeD increases the split DNN inference request success rate by 50%, guaranteeing an inference success rate of at least 96%, confirming its ability to optimize both transmission efficiency and inference performance.