Road Disease Recognition Algorithm Based On Multi-channel Convolutional Neural Network

In recent years,with the development of the transportation industry and the continuous increase of highway mileage,the requirements for road maintenance have also increased.Designing a fast and accurate method for identifying road diseases that can be applied in a wide range has become an extremely urgent need in road maintenance work.Most of the traditional road disease automatic recognition algorithms have problems such as slow recognition speed,low accuracy,and poor robustness.Aiming at these problems and combining the advantages of convolutional neural network method in the field of image recognition,this paper proposed a road disease recognition algorithm based on multi-channel convolutional neural network.The end of a typical convolutional neural network uses fully connected layers to integrate local features to obtain high-level features of the image.In this paper,the convolutional neural network is used to extract local edge information in the road image.The end of the convolutional network is fine-tuned to adapt to the needs of the scenario.In the traditional method,the recognition window is cut and divided,which results in a limited recognition field of vision and prone to sudden misrecognition.This algorithm addresses the shortcomings by using the overall zoom method to obtain the characteristics of the recognition window,which expands the recognition field and effectively uses the pixel information around the window,so that the extension characteristics of the disease in the road image can be well restored.Most of the existing methods use relatively fixed methods for feature extraction,but the characteristics of road diseases are actually very complicated.When using a single method to identify multiple special disease areas,the accuracy is not high.To solve this problem,this algorithm uses a dual network structure.Use a separate network to enhance special disease areas.Aiming at the problem that some disease features are not obvious in the actual recognition process and the algorithm has poor recognition effect,this algorithm uses a two-channel structure to perform binary segmentation on the original gray image,so that the disease area is more easily extracted.This paper organized a large amount of data to train and identify the proposed algorithm.Experimental results show that the accuracy of the algorithm is significantly higher than the traditional method,especially for various special diseases.Through comparative experiments,it is verified that several innovative improvements of the algorithm have played a role in improving the recognition effect.The experiment proves that the recognition accuracy of this algorithm is close to the level of human recognition.The recognition speed can also meet the requirements in the autonomous driving scenario,which has positive significance for the further development of road disease automatic identification technology and autonomous driving technology.