Mineral resources are important industrial raw materials and the cornerstone of ensuring industrial production, especially metal ores. With the continuous development and progress of artificial intelligence technology, it is of great significance to apply artificial intelligence technology to mining. Computer vision technology, as a sensor that collects information like a human “eye”, is becoming increasingly important in ensuring mining safety, improving mining continuity, and reducing environmental interference through computer vision methods. In this context, this paper focuses on general problems of metal mineral resources, the sustainability of exploration, drilling and blasting, transport, personnel safety, and security. It describes the latest progress of computer vision technology in each link and summarizes and looks forward to the key technical methods. It also summarizes and looks ahead to the key technical methods in each area. The research results show that the application of computer-vision-related technologies in related links not only greatly improves production efficiency but also reduces environmental interference and the probability of production safety accidents, effectively ensuring sustainable mining. In the future, to achieve unmanned mining throughout the entire process, it will be necessary to combine computer vision technology with other specialties such as intelligent control and intelligent perception to achieve a technological breakthrough throughout the entire process.

Block caving is a cost-effective mining method that enables the highly efficient mining of thick and large ore bodies. During ore extraction in block caving operations, the ore-drawing roadways require especially high safety standards. However, the complex in situ stress conditions and cyclic loading from caved ore significantly deteriorate the stability of the surrounding rock. This makes rock mass control particularly challenging, such that it is crucial to study an effective method for maintaining the long-term stability of the roadways. This research proposes a comprehensive approach combining laboratory rock mechanics testing, numerical simulation, and field engineering validation to design effective support strategies for disturbance-affected roadways. Laboratory tests provide accurate mechanical parameters for the rock mass, the numerical simulations allow for the comprehensive analysis of deformation–failure mechanisms under disturbance conditions, and field validation ensures the reliability and practical applicability of the proposed support method. This study focuses on a −285 m ore-drawing roadway in the western section of the Yanqianshan Iron Mine. The in situ stress distribution was characterized through rock mechanics testing and acoustic emission monitoring. The propagation mechanisms of ore-drawing disturbance waves within the rock mass were analyzed, and numerical simulations revealed the deformation patterns and failure modes under dynamic disturbance, upon which the support scheme was designed. The results demonstrate that the designed bolt–mesh–shotcrete support scheme can effectively control surrounding rock deformation within 5 mm and resists the deformation induced by cyclic disturbances. This study provides valuable technical support for stability management in block caving mines with similar conditions.

Accurate prediction of flotation concentrate grade is essential for maintaining product quality and supporting timely process adjustment. However, mechanistic prediction remains difficult because industrial flotation involves nonlinear multivariable interactions and partially observed operating states. Recent advances in machine learning have increased interest in data-driven methods, with encouraging results. Although recent studies have incorporated temporal dependencies into flotation-grade prediction, explicitly modeling the condition-dependent evolution of multivariate process–quality relationships under changing plant conditions remains challenging. In this study, industrial flotation records were systematically analyzed from a data-driven perspective to identify challenges affecting concentrate-grade prediction. Statistical analyses showed that variable–grade relationships are condition-dependent and vary across operating periods, indicating that measured variables only partially characterize evolving flotation states and that fixed input–output mappings may be inadequate. Accordingly, a DGraFormer-based model was developed to capture evolving inter-variable dependencies and multi-scale temporal patterns. For one-hour-ahead forecasting, the model achieved an RMSE of 0.7587 and an $R^2$ of 0.5423 for iron concentrate grade, and an RMSE of 0.6917 and an $R^2$ of 0.6384 for silica concentrate grade, outperforming the evaluated machine-learning and deep-learning baselines overall. These results underscore the importance of dynamic multivariate modeling for accurate concentrate-grade forecasting.

During production, copper mining could generate substantial waste rock that impacts land use and the environment. Advances in deep learning have enabled efficient, cost-effective intelligent sorting, where vision sensor performance critically determines sorting accuracy and efficiency. However, the sorting environment of copper mine waste rock is inherently complex, particularly within the conveyor belt section of the sorting machine, where insufficient and uneven lighting significantly impairs the performance of vision-based detection systems. To address the above challenges, a deep-learning-based copper mine waste rock detection algorithm under low-light environments is proposed. Firstly, an Illumination Adaptive Transformer (IAT) module is added as a preprocessing layer at the beginning of the Backbone to enhance the brightness of the images acquired by the vision sensor. Secondly, a Local Enhancement-Global Modulation (LEGM) module is integrated after the A2C2f and C3k2 modules in the Neck to enhance the detection accuracy. Finally, to further improve the model performance, MPDIoU is introduced to optimize the original loss function CIoU. As a result, the proposed algorithm achieved an mAP@0.5 of 0.957 and an mAP@0.5:0.95 of 0.689, outperforming advanced methods by 1.9% and 8.6%, respectively.