The existing various types of cabin cleaning machines are sufficient in terms of transportation capacity, endurance, and cleaning efficiency, but there are still some small problems. This article can focus on addressing the shortcomings of cabin cleaning machines in obstacle recognition and avoidance, target detection and positioning, enabling cabin cleaning machines to have certain intelligent decision-making capabilities and improving the efficiency of clearance work. This article uses the Residual Neural Network (ResNet) from Convolutional Neural Network (CNN) to build a system. After data collection, image denoising and adjustment, model training, and integration into autonomous perception and intelligent decision-making systems, the final model is integrated into the cabin cleaning machines. Finally, other algorithm models are applied for comparative experiments to verify the superiority and robustness of the proposed model. In the first set of experiments, the recognition accuracy of the model established by the algorithm in this article was between 0.9111 and 0.9428 during daytime, nighttime, sunny and rainy days, and the decision accuracy was between 0.9532 and 0.9797. The average recognition accuracy of the second group of experiments was between 0.9311 and 0.9502 during clear day, no rain at night, rainy day, and rainy night, and the average decision accuracy was between 0.9452 and 0.9841. The other four algorithms generally have accuracies below 0.9 or even 0.8 in both cases, and the fluctuation in accuracy is greater than that of the algorithm proposed in this article. The autonomous perception and intelligent decision-making system of the cabin cleaning machine based on convolutional neural network optimizes the cleaning process, reduces manual operation requirements, and successfully improves the perception and cleaning efficiency of the cabin cleaning machine towards the cabin environment, with strong robustness.