Convolutional Neural Network for Smoke Image Super-Resolution

Abstract

The1 flare stack is the last line of defense in the safe production of large-scale chemical plants. Monitoring black smoke produced by the incomplete flare stack exhaust combustion can effectively reduce environmental pollution and production accident. In order to improve the ability to recognition and analyze the black smoke, high-resolution flare stack scene images are in urgent need. To this end, we in this paper propose a super-resolution algorithm based on convolutional network that focuses only on smoke area for the purpose of identifying the smoke of flare stack. With a lightweight convolutional neural network structure, our network specializes in learning smoke characteristics mapping between the low-resolution images and the associated high-resolution. To verify validity, our algorithm compares the super-resolution quality of the smoky region of the flare stack image with several classic super-resolution algorithms. The experimental results show that our algorithm is superior to the classical algorithms when applied to smoke images.