Research on Smoke Image Recognition: Multi-Scale Dilated Convolutional Neural Network and Entropy Function Threshold

For the application of artificial intelligence in image-based smoke detection in emergency management, this paper proposes an image recognition algorithm based on a multi-scale dilated convolutional neural network and entropy threshold function to prevent the loss of small-scale features, low detection rate, and poor real-time performance of smoke image recognition. First, the smoke features are determined according to a multi-scale analysis model, and multi-scale information such as texture, color, and shape of the smoke image is obtained. Second, we propose a dilated convolutional neural network to solve the feature loss. Third, according to the characteristics of the entropy function, we build a model that comprehensively considers smoke features and recognition thresholds, which more accurately describes the image recognition process. Ultimately, this research develops a smoke detection system using the suggested algorithm and demonstrates its efficacy through simulation and experimental validation.