A Specific Task-oriented Semantic Image Communication System for substation patrol inspection

TL;DR

This work tackles the challenge of transmitting patrol-imagery from intelligent substation robots over weak wireless channels by formulating a task-oriented semantic communication system (STSCI) that prioritizes preserving key semantic content. It integrates a GAN-based auto-encoder with deep JSCC for robust transmission and introduces a semantic enhancement pathway using YOLONet to locate critical regions and an enhancement CNN to improve their fidelity, yielding improved image quality at low bitrates. Across COCO2014-derived data with substation fine-tuning and hardware-channel tests, STSCI demonstrates superior PSNR/SSIM performance compared to JPEG, JPEG2000, and LSCI, and approaches LDPC performance in favorable channels while maintaining robustness under low SNR. The approach offers practical potential for reliable, real-time inspection by patrol robots in weak-signal substations and can be adapted to other fixed-source/condition tasks by re-targeting semantic content during training.

Abstract

Intelligent inspection robots are widely used in substation patrol inspection, which can help check potential safety hazards by patrolling the substation and sending back scene images. However, when patrolling some marginal areas with weak signal, the scene images cannot be sucessfully transmissted to be used for hidden danger elimination, which greatly reduces the quality of robots'daily work. To solve such problem, a Specific Task-oriented Semantic Communication System for Imag-STSCI is designed, which involves the semantic features extraction, transmission, restoration and enhancement to get clearer images sent by intelligent robots under weak signals. Inspired by that only some specific details of the image are needed in such substation patrol inspection task, we proposed a new paradigm of semantic enhancement in such specific task to ensure the clarity of key semantic information when facing a lower bit rate or a low signal-to-noise ratio situation. Across the reality-based simulation, experiments show our STSCI can generally surpass traditional image-compression-based and channel-codingbased or other semantic communication system in the substation patrol inspection task with a lower bit rate even under a low signal-to-noise ratio situation.

Figures (10)

• Figure 1: The framework of STSCI. STSCI consists of two systems, the base system and the semantic enhancement system. The base system consists of a semantic encoder, a semantic decoder, and a simulated-channel model (trained only), with the process indicated by black lines. The semantic enhancement system with the process indicated by red lines, on the other hand, includes a YOLONet for identifying key semantic content and an enhancement CNN network that utilizes extra information to enhance the transmission quality of the key semantic information. The simulated-channel model is only used during the model training process to simulate a real-world wireless channel. This process is indicated by blue lines.
• Figure 2: The architecture of the base system, including the encoder at transmitting end and the decoder at receiving end. The modules with green background describes the process of image transmission. During training, a simulated-channel model is added for source-channel joint encoding, as well as a discriminator in the GAN model to assist in training the generator, which is the decoder. The networks in the diagram are represented by several rectangular blocks, with the number of blocks matching the number of network layers. Blocks of different colors represent different network layers.
• Figure 3: The process of the semantic enhancement system. The YOLONet is used to locate areas in the image that contain key semantic information. Those areas are then cropped into sub-images and transmitted to the receiving end through the basic system. This portion, serving as additional information, is combined with the complete image transmitted from the basic system and input into the enhancement CNN model to obtain the final image.
• Figure 4: The performance of the reconstructed image of JPEG, JPEG2000, LSCI and STSCI with different bpp. The metrics SSIM and PSNR(dB) measure the distortion performance.
• Figure 5: Visual example of images produced by LSCI along with the corresponding results for JPEG and JPEG2000. STSCI has higher metric value in a lower bpp than JPEG and JPEG2000. The image compressed by STSCI is more realistic and the details are clearer.