Aerial Photographic Information Recognition Based On Convolutional Neural Network

With the development of neural network technology and the improvement of image resolution,the information recognition and recognition technology of various images have attracted more and more attention.In recent years,the rapid development of Unmanned Aerial Vehicle(UAV)Aerial photography technology has put forward higher standards for information identification algorithm and classification.Traditional identification algorithm needs to design appropriate feature extraction tools for specific data sets.Compared with the traditional recognition algorithm,convolutional neural network can automatically recognize the complex structure of imported data,and its deep structure characteristics can integrate the simple features extracted from the shallow layer into the abstract features more suitable for the task requirements.By adjusting the number of layers and width of the structure,the learning ability of the convolutional neural network can be changed,so that the overall structure has enough flexibility.Aiming at the need of aerial photo processing,this paper proposes an aerial photo information recognition method based on convolution neural network.The experiment in this paper is to develop and improve the information recognition model based on regional convolutional neural network based on Tensor Flow machine learning platform to realize the recognition of target information based on aerial photography.First,the platform environment was built and the data set was prepared.Combined with the high flexibility and performance optimization of Tensor Flow platform,the high-efficiency GPU equipped laid a foundation for the experimental environment.Convolution neural network model contains a large number of free parameters,in the case of less training data,it's easy to have a fitting problem,through the way of data expansion can not only effectively expand the training sample number,you can also avoid happened fitting problem,on the one hand,increase the diversity of the training sample,on the other hand also can bring model performance boost.Secondly,the structure and working principle of convolutional neural network in information recognition are known.And a representative regional convolutional neural network in the field of information recognition R-CNN,Fast regional convolutional neural network Fast R-CNN,Faster region convolutional neural network Faster R-CNN,You only watch the Internet once YOLO,Single detector algorithm SDD study separately,Through the research and comparison of each network structure,the construction principle of each network can be intuitively understood.In combination with existing experimental conditions and multi-directional factors,optimization and improvement of the YOLO model can be selected.Finally,the network structure of the YOLO model with the optimized parameters was adjusted,including 24 convolutional layers,4 pooling layers and 2 fully connected layers.The multi-layer structure ensures the recognition ability of the model.The experiment USES Image Net to verify the vegetation images on the data set,and the accuracy of the pre-training of the model reaches 88%,after which the model is used to process the aerial photographic data set and achieve the expected effect.The experiment of aerial photography has proved the effectiveness of this method.The results show that the method of information recognition studied in this paper has high accuracy,fast speed and low labor cost.