Research On Slope Disease Identification Based On Convolutional Neural Network

The hydrological situation in China has always had significant seasonal differences,with rainy and waterlogged summers and relatively dry winters.The northern region often faces water resource shortages,while the southern region has relatively abundant water resources.In the past twenty years,China has successively built a number of major water diversion projects,and the slope structures of these projects are susceptible to safety hazards caused by internal aging,external environmental factors,and other factors.Slope safety is not only directly related to the smooth operation of water conservancy projects,but also to the safety of the lives and property of the surrounding people.However,traditional survey methods often have problems such as high labor costs and low identification efficiency.Therefore,a reliable and efficient system for identifying slope diseases has great engineering application value.This article studies and designs a slope disease image recognition system based on convolutional neural networks.Firstly,drones and handheld devices are used to collect images of river and waterway slope diseases and create a dataset.Retinex algorithm is used to enhance the dataset,and augmentation operations such as flipping,scaling,and color gamut changes are used to expand the limited initial dataset,laying the foundation for subsequent experiments.Secondly,in-depth research and analysis were conducted on the YOLOX network model,and a slope disease identification and detection algorithm was designed using YOLOX-S.On the basis of the original network model,the confidence function is replaced by Focal loss function in order to alleviate the problem of unbalanced sample number,and ECANet attention mechanism is introduced to make the model pay more attention to the given disease characteristics.Finally,a slope disease identification application system was proposed and designed,which has functions such as disease type detection,disease localization,and historical disease comparison.The disease detection module,as the core content,can classify and mark the incoming video images for diseases.The location positioning module obtains the longitude and latitude information of the detected disease location.The disease comparison module can view historical detection information and compare it with historical results.The experimental results show that after improving YOLOX-S model algorithm,the mean average precision of crack,defect,landslide diseases on slopes reached 96.9%,effectively improving model detection accuracy.The slope disease recognition system developed based on this model can also provide help for relevant professional personnel use.