Aerial Image Visual Detection And Recognition Method Of Ground Crack In Goaf Of Coal Mine

After the completion of coal mining,the remained goaf region in coal will collapse after a period of time,resulting in cracks in the land above the surface,seriously damaging the local environment and buildings,and threatening life safety.Periodically inspection of ground crack over mine goaf is a necessary task for environmental protection and human safety.However,the traditional manual inspection method is time-consuming,subjective and with potential dangers because the goaf normally locates in remote regions and inevitably induces complicated and gullies-cross landform.Therefore,Unmanned Aerial Vehicle(UAV)with high resolution camera is adopted instead of manual to capture aerial images in goaf,then image processing and deep learning method are utilized to automatically detect and recognize ground crack.At the same time,detection system of ground crack is built,and the corresponding crack risk assessment method is proposed,which provides certain basis for the subsequent measures and loss accounting.The main work is as follows.(1)Aerial image detection method of ground crack based on deformable convolutional network with hybrid domain attention is proposedIn view of long and narrow characteristics of ground crack in aerial view,deformable convolution network with hybrid domain attention is constructed to detect ground crack.ResNet is utilized as backbone,and structure of feature pyramid network is used to enhance the feature information of each stage.Considering that the standard convolution can not accurately obtain the information of receptive field of ground crack,the deformable convolution is employed instead for feature extraction.At the same time,hybrid domain attention mechanism is proposed to strengthen channel and spatial information of feature maps,where channel domain attention mechanism is used to set different weights for each channel of feature maps,which strengthen the contributions of specific channels.Spatial attention mechanism is used to set different weights for each element in feature maps,which strengthen the contributions of specific spatial locations.Based on it,the attention guided common detection framework of ground crack is proposed in the thesis,UAV with high-resolution is employed to capture images and detect cracks.The experimental results show that the proposed methods are more stable to training and has better detection accuracy compared to other detection model.(2)Ground crack recognition method based on fully convolutional network with multi-scale input is proposedDifferent from predicting bounding box by detection method,semantic segmentation method can achieve pixel-level recognizing accuracy.Ground crack only distribute in a very limited part in aerial images,which leads to imbalance problem.And there are complicated noises sharing similar characteristics with crack,such as shadow,cliff,terrace edge.To overcome the obstacles,ground crack recognition method based on fully convolutional network with multi-scale input is proposed.Firstly,a statistic pre-processing method is employed to remove useless patches before training,with the purpose of improving the training efficiency.Following that,fully convolutional network with multi-scale input semantic segmentation model is built,where multi-scale input method is applied to enrich the semantic and detailed information of feature maps at each stage.Furthermore,a multi-scale connection module is constructed to select the most important contextual information from the features under multiple scales.Finally,multi-stage features are used for recovering high-dimension features,and output the prediction result.The experimental results show that the proposed method has better performance than other crack recognition methods.In the experiments of sensitive analysis,the effectiveness of the proposed modules is proved by experiments.(3)Detection system of ground crack in goaf is built,and risk assessment method of ground crack is proposed.In order to meet the needs of mine digital and information development,the detection system of ground crack in goaf is designed,which includes data acquisition system and data management system.Between them,the data acquisition system used UAV,ground station and high-resolution pan tilt camera to collect aerial images of each goaf.The data management system employed the B/S architecture,includes three parts: data upload,ground crack detection and history query and statistics.In addition,the risk assessment method of ground crack is proposed.The skeleton information of ground crack is extracted by image processing method,the length,width,area and other geometric characteristics of ground crack are calculated according to the skeleton and crack contour information.And the corresponding risk assessment is given according to the geometric characteristics of the crack.In this thesis,the detection and recognition of ground crack are realized,and the detection system of ground crack is constructed,and the corresponding risk assessment method is given,which has certain guiding significance for the inspection and subsequent measures of ground crack in goaf of coal mine.The thesis includes 46 figures,15 tables and 101 references.