煤炭智能开采是煤炭行业高质量发展的必由之路，放顶煤工作面的智能化滞后于综采工作面。智能放煤是实现智能放顶煤开采的关键核心技术，低照度、小空间、高粉尘、煤矸叠压、声振信号干扰、夹矸误识别等问题严重制约智能放煤技术开发。在探索尝试图像、声音、振动等多种煤矸识别技术的基础上，提出了图像识别智能放煤技术。精准快速识别混矸率与适应恶劣环境是图像识别智能放煤需要攻克的主要技术难题。针对混矸率识别问题，将混矸率细化为投影面积混矸率、表面体积混矸率、内部体积混矸率。建立了轻量级的放顶煤工作面矸石识别及边界测量模型，实现了投影面积混矸率的精准快速识别。提出了快速写意重建和精准重建等2种煤矸块体三维重建方法，研究了煤矸块体三维形态特征与二维形态特征关系，揭示了后部刮板输送机上煤矸块体堆积特征。在此基础上，提出了“由表及里”的混矸率高精度预测2步走策略，以实现透明化煤流，达到混矸率高精度测量的目的。针对低照度、高粉尘等恶劣环境适应问题，提出了立体视觉照度智能监测与自适应调节方法，为图像采集实时提供最优照度环境。基于人体仿生学以及边缘AI技术开发了智能图像采集系统，提出了基于频域先验的单通道Retinex去粉尘算法以及空频域联合强化去粉尘算法，为图像识别持续提供高质量图像。形成了“三位一体”夹矸智能识别技术，对放煤过程中可能出现的夹矸进行了精准识别，减少了由于夹矸放出而引发的误识别、误操作。图像识别智能放煤技术可以提高放煤工序的智能化水平，提高资源回收率、降低含矸率，保证矿井安全生产，研究成果的科学应用将有助于高质量实现智能化放顶煤开采。

Intelligent coal mining is an inevitable course for the high quality development of the coal industry. The intellectualization development of longwall top coal caving (LTCC) face lags behind the fully mechanized face or heading face. The intelligent top coal caving is the core technology of achieving an intelligent top coal caving mining. However, there are some problems, such as low illuminance, narrow space, high concentration dust, overlapping coal and rock, interference of acoustic vibration signal and rock parting, which seriously restrict the development of intelligent top coal caving. On the basis of exploring a variety of recognition methods based on image, sound and vibration, an image based recognition of withdrawn coal in LTCC face (IB-LTCC) was developed. Accurate and rapid recognition of rock mixed ratio and harsh environment adaption are two main technical problems that need to be solved for IB-LTCC. In order to solve the former problem, the rock mixed ratio was refined into projection area rock mixed ratio, surface volume rock mixed ratio and inner volume rock mixed ratio. The lightweight rock recognition and boundary measurement model used for LTCC face was established, which achieved an accurate and rapid recognition of projection area rock mixed ratio. Two types of three dimensional blocks reconstruction methods including fast freehand brushwork reconstruction and accurate reconstruction were put forward. The relationship between three dimensional morphological characteristics and two-dimensional morphological characteristics of coal and rock blocks was studied, and the packing mechanics of coal and rock blocks on the rear armored face conveyor (RAFC) were revealed. “Surface to Inside” (S2I) protocol to measure rock mixed ratio (RMR) with high accuracy was put forward, so as to realize the transparency of coal flow and achieve the purpose of high precision measurement of rock mixed ratio. On the other hand, aiming at the problem of adaptation to harsh environment such as low illuminance and high concentration dust, stereo vision based illuminance measurement used for intelligent lighting was proposed to provide the optimal lighting environment for image acquisition in real time. The intelligent image collection system was developed based on human bionics and edge AI technology. Single channel Retinex dust removal algorithm based on frequency domain prior and an enhanced dust removal algorithm based on space frequency domain were proposed to continuously provide some high-quality images for image recognition. The“three in one”rock parting intelligent recognition technology was developed, which accurately recognized the rock parting that may appear in the coal caving process, reduce the mis recognition and mis operation caused by the drawing of rock parting. The IB-LTCC can improve the intelligent level of coal caving process, improve resource recovery, decrease rock mixed ratio and ensure the safe production of coal mine. The scientific application of research achievement will be helpful for achieving an intelligent LTCC with high quality.

1 煤矸图像特征的照度因素影响机制与煤矸识别特征选取

2 煤矸图像精准分割与混矸率计算

3 基于立体视觉的照度智能监测与自适应调节

4 顶煤放出规律与放出煤矸块体堆积特征

5 基于二维图像的煤矸块体形状表征模型与三维快速写意重建

6 基于多视图像序列的煤矸块体三维精准重建与体积测量

7 煤矸块体三维形态特征与二维形态特征关系

8 夹矸识别与异常放煤状态甄别

9 基于边缘AI的图像识别智能放煤关键设备开发

10 结论