With the continuous progress of technology, the equipment and technical complexity of thermal systems are increasing, resulting in higher requirements for thermal inspection systems. Traditional inspection systems often have problems such as un-reasonable path planning and incomplete equipment monitoring, which may lead to low equipment operation efficiency and even lead to safety accidents. To address these issues, this study proposed using chaotic optimization algorithms to optimize Proportion-integral-derivative (PID) controllers and using Resilient Back Propagation (RPROP) to optimize image denoising models. Subsequently, a new thermal inspection system was constructed based on improved PID controllers and optimized image denoising models. In this study, the thermal system data in the public database and the problem data set collected in the actual thermal environment were used for experimental verification. Through comparative experiments of denoising models, the improved image denoising model had a CPU running time of 0.35 s, 0.78 s, and 0.69 s for Parrot, Phantom, and Frame denoising, respectively, which was superior to the comparison models. Later, in the comparison experiment of improving the PID controller, the adjustment time, peak time, and rise time of the chaos optimized PID controller were 0.719 s, 0.731 s, and 0.595 s, respectively, which were superior to the comparative controller. Finally, an empirical analysis was conducted on the proposed new thermal inspection system. The average detection accuracy and average detection time of the new thermal inspection system were 94.6% and 28.4 minutes, respectively, which were significantly better than traditional thermal inspection systems. The above results indicate that the proposed new thermal inspection system has good performance and can be applied to the actual inspection process of thermal equipment, thereby promoting the development of the thermal systems.