我国厚煤层储量丰富，厚煤层高效开采对保障我国能源持续性供给意义重大。综放开采是我国独具特色的厚煤层高效开采方法，放煤动作的精准控制及决策是实现智能综放开采的核心难点。但由于缺乏有效的技术手段，现阶段综放开采放煤控制仍以人工放煤为主，严重影响厚煤层开采效率，亟需创新理论并研发新方法来降低综放工作面工人劳动强度。论文以曹家滩122107综采工作面为研究对象，采用现场调查、理论分析、放煤实验和验证分析等方法开展综放工作面顶煤放落影像监测特征与放出规律研究，主要工作如下：

分析了曹家滩煤矿工程地质条件及煤层赋存特征，确定了制约厚煤层综放开采放煤动作精准控制的影响因素；基于放煤特征及影响因素，自主研发了顶煤放落模拟实验平台，构建了粉尘环境下煤矸混合放出状态模拟实验，设计了暗通道去雾的粉尘环境优化算法和运动模糊的煤矸图像增强算法，提出了基于顶煤放落智能监测自适应调节方法，解决了顶煤放落智能监测影响因素的问题；提出了一种区域卷积神经网络的煤矸混合分析识别方法和基于双光流网络融合卷积神经网络的顶煤放落影像实时监测识别方法，系统揭示了顶煤放落过程的瞬态特征及影像特性；此外，总结了顶煤放落时上覆岩层松散区域凹陷表面积变化趋势，明确了基于时间及状态属性的顶煤层松散区域凹陷表面积变化规律及顶煤放落智能监测影像特征与放出规律间关系，研究表明，顶煤放落过程上覆岩层松散区域表面积随时序的变化呈现快速增长直至出现渐进稳定效应；基于多种方案的顶煤放落实验，顶煤放出率与检测平均准确率变化呈正相关，含矸率与平均漏检率变化趋势呈正相关，这为放煤动作的精准控制决策及实现透明化放煤提供了可靠的依据。

本研究为智能化综放开采放煤精准决策提供了理论支撑。

China has abundant reserves of thick coal seams, and the efficient mining of thick coal seams is of great significance to ensure the sustainability of China's energy supply. Fully mechanized caving mining is a unique and efficient mining method for thick coal seams in China. The precise control and decision of coal release action is the core difficulty to realize intelligent comprehensive release mining. However, due to the lack of effective technical means, the control of coal release in header mining is still dominated by manual coal release, which seriously affects the mining efficiency of thick coal seams. The paper takes Caojiatan 122107 top coal release working face as the research object, and uses field investigation, theoretical analysis, coal release experiment and verification analysis to carry out the research on the monitoring characteristics and release law of top coal release image of the comprehensive release working face, the main work is as follows:

Statistical analysis of the engineering and geological conditions and coal seam inventory characteristics of Caojiatan coal mine, determination of the influencing factors that restrict the accurate control and decision of coal release action of comprehensive coal mining in thick coal seams; based on the coal release characteristics and influencing factors, independent research and development of the top coal release simulation experiment platform, construction of the coal gangue mixed release state simulation experiment under dusty environment, design of the dusty environment optimization algorithm of dark channel defogging and motion blur coal gangue Image enhancement algorithm, proposed an adaptive adjustment method based on top coal release intelligent monitoring, and solved the problem of "internal-external" influencing factors of top coal release intelligent monitoring; proposed an improved real-time regional convolutional neural network for coal gangue mixed state analysis and identification method based on dual optical flow networks fusion deep convolutional neural network-based top coal fall image real-time monitoring identification method, systematically revealing the transient characteristics and image characteristics of the top coal fall process; In addition, summarize the change trend of depression surface area in the overlying rock loose area during the top coal fall, clarify the change law of depression surface area in the top coal loose area based on time and state attributes and the relationship between the top coal fall intelligent monitoring image characteristics The relationship between the top coal release rate and the average accuracy of detection is positively correlated; the gangue rate and the average miss detection rate are positively correlated; the study shows that the top coal release rate and the average accuracy of detection are positively correlated based on the top coal release experiments of various schemes; This study provides technical reference for intelligent fully mechanized caving mining and intelligent monitoring of coal drawing in fully mechanized caving face, and provides theoretical support for transparent coal drawing.

This study provides technical and theoretical support for accurate control and decision making of coal release in intelligent comprehensive discharge mining.