Carbonate clay-type lithium ore holds significant potential due to its extensive reserves, broad distribution, and relatively easy extraction. However, it presents significant beneficiation challenges due to its coexistence with karst-type bauxite, which often results in mixed, low-lithium ores. For the first time, spectral–texture features derived from hyperspectral imaging (HSI) are jointly modeled with a deep learning framework to explore the feasibility of pre-sorting carbonate clay-type lithium ores. Initially, the spectral responses reflecting mineral composition and the texture features characterizing structural differences were analyzed to evaluate the feasibility of using HSI for ore sorting. Furthermore, the influence of band selection, data standardization, and water absorption regions on pre-sorting performance was systematically investigated through comparative analysis of multiple dataset configurations. Two classification schemes, primary ore types classification and multi grade classification, were employed to assess sorting accuracy and identify key influencing factors. The proposed model, A2ST-OSNet, achieves excellent results in both ore localization and classification through staged data input with varying dimensions and a modular design. Results revealed that joint modeling of spectral and texture features enables efficient and accurate pre-sorting, whereas models relying solely on either spectral or texture features were insufficient, as discriminative information for ore sorting is jointly determined by mineral composition and structural characteristics. Moreover, refined feature extraction and fusion strategies, including spectral–texture feature selection and attention mechanisms, proved critical in enhancing classification performance. The proposed approach offers valuable technical support for ore beneficiation and tailings reutilization, contributing to sustainable resource utilization and providing an effective solution for the efficient recycling of carbonate clay-type lithium ores.
扫码关注我们
求助内容:
应助结果提醒方式:
