Abstract: The main task of image-based intelligent control of caving door in longwall top coal caving face is to monitor the rock mixed ratio of coal flow by analyzing the image, and then determine the closing time of caving door. However, the three-dimensional shape of coal rock particle and the accumulation information of coal flow cannot be directly obtained by two-dimensional image. Based on the recognition of rock mixed ratio on the surface of coal flow, aiming at the recognition of inside rock mixed ratio of the coal flow, a novel method of site-directed mutagenesis (SDM) for morphological genes of coal and rock particles was proposed. The morphological characteristics of coal rock particles under different mutagenesis sites and different mutagenesis intensities were studied. The relationship between morphological gene sites of coal rock particles and macro-, meso- and micro-morphological characteristics of coal rock particles was quantified, and the feasibility of rapid batch generation of coal rock particle cluster with expected morphological characteristics distribution was explored. The accumulation characteristics of coal rock particles under the disturbance of rear armored face conveyor (rear AFC) are studied, and the mapping relationship between the surface rock mixed ratio and the inside rock mixed ratio coal flow is determined. Considering the irregular shape of coal rock particles and the stacking pressure of coal flow, a prediction model of cumulative volume-based rock mixed ratio based on interval volume-based rock mixed ratio is established. The research shows that the morphological characteristics of different scales of coal rock particles are stored in different sites of morphological genes. The larger the morphological gene sites, the finer the morphological characteristics stored. The morphological genes are further divided into macro-, meso- and micro-morphological genes. By using morphological primers to site-directed mutagenesis on the morphological genes of coal and rock particles to realize the precise modification and controllable variation of the morphological characteristics of the specific scale of the particle, which can be used to quickly generate the coal and rock particle cluster with the expected morphological characteristics distribution. The appropriate image acquisition position of image-based intelligent control of caving door is determined, and the traction and disturbance of the rear AFC are utilized to change the accumulation characteristics of the coal flow and improve the prediction accuracy of the internal volume-based rock mixed ratio of the coal flow. Using the interval rock mixed ratio instead of the instantaneous rock mixed ratio as the basis for closing the caving door can increase the drawing time and improve the coal recovery rate without significantly increasing the rock mixed ratio. The determination coefficient of the cumulative volume-based rock mixed ratio prediction model is 0.9783, which realizes the reasonable prediction of the internal volume-based rock mixed ratio of the coal flow by using the two-dimensional image. A strategy to ensure the prediction accuracy of the internal rock mixed ratio of coal flow based on dual-energy X-ray transmission technology was further proposed, which provides a novel approach to guarantee the prediction accuracy of rock mixed ratio under complex conditions.