Longwall Top Coal Caving (LTCC) mining technology is an effective method for mining thick and extra thick coal seams in China.Since its application in China in 1982, significant progress has been made, forming a variety of mining modes such as LTCC with large mining height, LTCC in extra thick coal seams and LTCC in steeply inclined thick coal seams. Caving door is an essential part of the support in LTCC face, which determines the beginning and end of coal caving. But the control of caving door is still high labor intensity and low working efficiency. There are technical problems in intelligent control of the caving door, such as difficulty in adapting to the complex underground environment, monitoring the migration state of top coal, and estimating the rock mixed ratio. Based on BBR theory, an image-based intelligent control technology of caving doorwasintroducedinthispaper.Thethird-generationtopcoalmigration tracing marker is developed to realize multiple rounds of coal caving. The effect of illuminance on the image features of coal and rock is analyzed. An improved Retinex enhancement algorithm is proposed to remove dust and fog. A lightweight multi-scale rock boundary detection model and the rock volume estimation model are established to estimate the rock mixed ratio in real-time. The hardware and software of image-based intelligent caving door control technology are developed and tested in both the laboratory and in-site. The results show that the features between coal and rock can be enlarged when considering optimal illuminance. The dust and fog can be effectively removed using the proposed algorithm and the developed hardware. The research results have been successfully applied in many coal mines, and the working efficiency and top coal recovery ratio have been improved.