机器视觉灰度化金字塔卷积模型的煤流异物识别
Recognition of unwanted objects in coal flow based on gray pyramid convolution model of machine vision
-
摘要:
煤流的带式输送是煤炭能源生产过程中的一个重要环节，煤流中异物识别是智慧矿山智能矿井建设中不可或缺的重要任务。针对煤流的带式输送监测系统中异物识别问题，构造一种基于机器视觉灰度化的双路金字塔卷积识别模型。首先，提出一种基于权值训练的机器视觉图像灰度化方法，并根据输送带运输异物与煤的颜色差异和运动性差异，采用Lucas-Kanade光流算法，实现对异常颜色和异常运动像素点的识别。同时结合光照补偿处理，实现对煤流实时图像进行亮度校正和无关像素剔除的预处理。接着，构造了金字塔网络，提取煤流中移动异物的多尺度融合特征。之后本文构造图像的常规特征提取通路和特征差分通路的双通道网络模型，实现对煤和异物特征的识别。最后，对陕煤曹家滩煤矿的煤流输送带监测图像的异物进行识别。结果表明，基于机器视觉灰度化双通道金字塔卷积模型对煤流输送带的异物识别准确率为95.7%，比经典的卷积网络模型的识别精度提高了2.1%，相比于传统图像灰度化固定权值识别模型，基于权值自适应训练的异物识别模型精度提升了5.2%。
Abstract:The belt conveying of coal flow is an important link in the process of coal energy production, and the recognition of foreign matters in coal flow is an indispensable task in the construction of intelligent mining. In view of the problem of the unwanted object recognition of the coal flow in belt conveyor monitoring system, a dual pyramid convolution recognition model was constructed by using the machine vision grayscale. Firstly, a method of machine vision image graying based on weight training was proposed. According to the color difference and motion difference between foreign matters and coal transported by conveyor belt, Lucas-Kanade optical flow algorithm was used to realize the recognition of abnormal colors and abnormal motion pixels. At the same time, combined with illumination compensation processing, the real-time image of coal flow is preprocessed by brightness correction and irrelevant pixel removal. Secondly, a pyramid network was constructed to extract the multi-scale fusion characteristics of moving foreign matters in coal flow. This paper constructed a two-channel network model of conventional feature extraction pathway and feature difference pathway to realize the recognition of coal and foreign matter features. Finally, the foreign matters in the monitoring images of coal flow conveyor belt in Caojiatan coal mine of Shaanxi coalfield were identified. The results show that the foreign body recognition accuracy of coal conveyor belt based on machine vision grayscale two-channel pyramid convolution model is 95.7%, which is 2.1% higher than that of classical convolution network model. Compared with traditional image gray-scale fixed weight recognition model, the accuracy of foreign object recognition model based on adaptive weight training is improved by 5.2%.
-
-
期刊类型引用(6)
1. 侯伟，黄金堤，李明周，李静. 基于YOLOv8算法的稀土熔盐电解槽炉面温度监测研究. 有色金属(冶炼部分). 2025(01): 84-91 . 
百度学术
2. 任建超，张虎. 基于机器视觉的煤矿带式输送机智能调速系统研究. 煤矿机械. 2025(02): 193-195 . 百度学术
3. 张炳建，郝洪涛，刘秀娟，李泽旭. 基于改进YOLOv5s的煤矿输送带异物检测算法. 宁夏工程技术. 2024(02): 174-179 . 百度学术
4. 梅晓虎，吕小强，雷萌. 基于Stair-YOLOv7-tiny的煤矿井下输送带异物检测. 工矿自动化. 2024(08): 99-104+111 . 百度学术
5. 王勇. 浮选产品煤质快速分析检测技术的应用. 山西化工. 2024(11): 64-66+92 . 百度学术
6. 蒋社想，周馨蕊. 带式输送机智能巡检系统设计. 煤炭技术. 2023(05): 203-206 . 百度学术
其他类型引用(5)
计量
- 文章访问数: 105
- HTML全文浏览量: 9
- PDF下载量: 170
- 被引次数: 11
下载:
