Pedestrian Detection Model in Underground Coal Mine Based on Active and Semi-supervised Learning

In the field of unmanned driving and safety monitoring of underground coal mine auxiliary transport vehicles, pedestrian detection technology is very important. At present, many works have carried out corresponding research on the influence of special working conditions such as dim light, uneven illumination, complex backgrounds, small and dense pedestrian targets in coal mines. However, these studies require the use of a large number of accurately labeled images, and it is difficult to obtain underground coal mine images, and the labeling is also very difficult. These factors have greatly affected the application and promotion of related models. To solve this problem, this paper proposes a pedestrian detection model in coal mines based on active learning and semi-supervised learning. Active learning can select indistinguishable data from existing unlabeled data for a small amount of labeling, while semi-supervised learning can use a small number of labeled samples and a large number of unlabeled samples to train the detection model and can achieve the same level as training with a large amount of labeled data. The detection accuracy of the model is similar. Therefore, the use of active learning and semi-supervised learning frameworks can effectively reduce the large demand for high-quality labeled data for the pedestrian detection model in coal mines. Experiments show that the coal mine underground pedestrian detection model based on braking learning and semi-supervised learning can use only 5% of the original data set for training to obtain a more effective detection model, which significantly reduces the dependence on labeled data and helps the rapid application and promotion of the pedestrian detection model in underground coal mines.