Research On Image Defogging And Target Recognition In Coal Mine

The underground image and video monitoring system is one of the important safety guarantees in the mine field.When the equipment runs abnormally or occur the personnel operation error,it can remind the staff to correct the hidden dangers and prevent the accident in time by sending a warning.Due to the complex underground environment of coal mines,dust or residual fog after dust removal often occurs in the environment.Many monitoring equipment are limited and cannot achieve normal functions.In order to overcome the influence of dust and fog characteristics on image information and improve the performance of the monitoring system,this paper studies the image defogging and pedestrian target detection algorithms on the basis of predecessors.The specific contents are as follows:1.Aiming at the problem that the brightness of the restored image is reduced due to the inaccurate atmospheric light intensity,this paper uses an atmospheric light intensity fusion estimation method.Firstly,the formation principle of atmospheric light intensity in the classical dark channel prior theory is analyzed,and based on this result,the bright channel prior theory is introduced to establish a new method for estimating local atmospheric light intensity in images.2.Aiming at the problem that initial transmittance edge optimization effect is unsatisfactory,this paper proposes a new refinement method of improved weighted guided filtering.Firstly,a regularization factor weight coefficient is constructed by using the difference between the pixel center point and the neighborhood point in the gray domain after analyzing the filtering characteristics of bilateral filtering.Then,for improving the image edge processing effect,it adds an adaptive adjustment function to adjust the regularization factor adaptively.3.In order to improve the brightness of the restored image furtherly,this paper proposes an adaptive adjustment function.The difference between the pixel point of the original dust-fog image and the atmospheric light intensity of the dark channel is selected as the adjustment factor to adjust the brightness and color of the image adaptively,then restore the fog-free image.4.Based on the Anchor-free network Centernet,this paper proposes a lightweight improved algorithm Centernet-MobileNet that uses MobileNet-V2 to replace Rensnet50 in the original model as the image feature extraction network.Backbone uses deep separable convolution and inverse residual structure,which can improve the efficiency of image target feature extraction and reduce the parameters of the model effectively.In order to improve the detection effect of small targets,enhance overall detection accuracy and convergence speed of the model,this paper adds the feature pyramid idea and attention mechanism.The experimental comparison results indicate that the improved image defogging algorithm proposed in this paper gets a good defogging effect,and improves the image restoration quality on the basis of ensuring the detection speed effectively.The improved pedestrian target detection algorithm proposed also has a great improvement in detection speed and detection accuracy.The number of weight parameters is only 1/4 of the original Centernet model,and the model achieves a better lightweight.