基于改进Faster-RCNN的矿工检测

Miner Detection Based on Improved Fast-er-RCNN

摘要: 针对Faster RCNN仅使用单一全连接层用于目标分类和边界框回归进而导致检测准确率不高的问题，提出了基于改进Faster RCNN的矿工检测框架，即将目标分类问题和边界框回归问题分开处理。为了实现矿工出入计数功能，在改进Faster RCNN框架的识别后，设计了基于三线的矿工出入检测算法。在真实的井下数据集上实验表明，改进Faster RCNN检测框架有更高的查准率和查全率。

Abstract: Aiming at the problem that the detection accuracy is not high when the single full connection layer is only used for target classification and bounding box regression by Faster RCNN, a miner detection framework based on improved Faster RCNN is proposed, which deals with target classification and bounding box regression separately. In order to realize the function of miner in and out counting, after improving the recognition of Faster RCNN framework, a miner in and out detection algorithm based on three lines was designed. Experiments on real downhole data sets show that the improved Faster RCNN detection framework has higher accuracy and recall.