Ash Content Detection in Coal Slime Flotation Tailings Based on Neural Architecture Search

Timely and accurate detection of the ash content value in tailings is crucial for guiding coal slime flotation production. Currently, there are two main methods for ash content detection: manual combustion of tailings samples and visual ash measurement based on deep learning. The method of manual combustion of tailings is time-consuming and difficult to provide prompt guidance for production. In contrast, the ash content detection method based on deep learning can significantly improve work efficiency. However, manually designing deep neural network models is time-consuming, requires high professional knowledge, and can easily lead to errors. To address this issue, the optimal network for tailings ash content detection is automatically generated using neural network architecture search based on reusable cell. The neural network obtained from the cell architecture search with the optimal performance is named NDRN. A dataset is created by collecting tailings images from coal flotation plants. NDRN and other deep learning models are trained and compared on this dataset to demonstrate the excellent accuracy and practicality of NDRN in tailings ash content detection tasks.