<p>This paper addresses the energy efficiency bottleneck caused by “full-hop recursive optimization” in multi-hop semantic communication systems by proposing an adaptive selective computation mechanism based on quality thresholds. The core innovation of this mechanism lies in establishing a “quality-aware dynamic decision-making” optimization paradigm: using the widely accepted quality benchmark (Peak Signal-to-Noise Ratio (PSNR) = 30dB) as the threshold, and employing Mean Squared Error (MSE) to quantify the distortion level at each hop. The mechanism evaluates the reconstructed semantic quality in real-time at each hop during multi-hop transmission, triggering the recursive optimization process only for samples that fail to meet the threshold, while skipping redundant computations and backpropagation for samples that already satisfy the quality requirements. Extensive experiments were conducted on the CIFAR-10 dataset using two representative semantic communication architectures, Vision Transformer-based Semantic Communication (ViTSC) and Deep Joint Source-Channel Coding (DeepJSCC), over both Additive White Gaussian Noise(AWGN) and Rician fading channels. Experimental results demonstrate that in the AWGN channel with 3 hops, the energy consumption and latency of ViTSC are reduced by 14.9% and 12.8%, respectively, while those of DeepJSCC are reduced by 8.1% and 7.5%, respectively. This method significantly reduces energy consumption while maintaining transmission quality. This work contributes a lightweight and effective optimization strategy for practical multi-hop semantic communication systems.</p>