Research On Vision Recognition Technology Of Forest Trails Based On Deep Neural Network

With the maturity of the research and development technology of Unmanned Aerial Vehicles(UAVs)and the manufacturing cost has been greatly reduced,UAVs have been widely used in lots of areas.It has been continuously developed in civil fields,such as environmental monitoring,geological exploration,forest fire prevention,and the applicable fields are still being expanded.In recent years,with the popularity of artificial intelligence technology,the combination of UAVs and artificial intelligence to improve the autonomy of UAVs has become a hot spot in autonomous navigation.It has achieved promising performances in a known or structured environment.However,the autonomous flight technology is still a challenge in the unstructured environment such as deserts or forests.The safest and most effective way of flying in a forested environment for UAVs is to detect and follow a forest trail.How to detect the forest trail accurately and quickly is the key to the study.Convolutional neural network(CNN)has achieved outstanding results in the field of image recognition and classification due to its strong generality and robustness.It has become one of the mainstream technologies in path recognition nowadays.Considering the difficulty of trail recognition in forests,a model of two-column Deep Neural Networks(2CDNN)was proposed.The proposed method judged the flight direction of UAV by detecting the forest footpath,ensuring that UAV keeps flying along the forest trail.Firstly,CIELab color space was transformed to retain more clear features,and then combined histogram equalization with edge extraction to obtain the feature map.The features were fed to a two-parallel deep residual network to extract color and texture features from forest images.UAV controlling command was generated according to classified results.The proposed method has been evaluated on IDSIA dataset and achieved accuracy of 91.31%,which delivers a 4.41% improvement than pervious methods.In addition,the method was also performed in the AirSim simulation environment for performance verification.It was showed that the flight trajectory simulated by the 2CDNN model was consistent with the forest trail.After fine-tuning the model with the synthetic training set collected in the AirSim environment,the flight trajectory can be closer to the forest trail,and the accuracy rate can be further improved.