| As the most widely used image recognition technology,depth learning plays a very important role in the field of remote sensing image processing as well as its successful application in artificial intelligence and computer vision.So,the traditional method of ground object classification is changed to the deep learning to realize the classification of ground object in the high resolution remote sensing image.Most of the previous studies have realized the extraction and classification of agricultural object information by using deep learning.This paper proposes a new method to recognize TOPA(typical objects of protected agriculture,TOPA)on the basis of using Tensorflow to build the model of deep learning and designing the recognition algorithm of TOPA.The TOPA are mainly including plastic greenhouse,multi-span greenhouse and sunlight greenhouse.Besides,the experimental area of protected agriculture in Huzhu,Qinghai province is taken as the research area.Through the deep learning of TOPA in the UAV image data,the accurate identification of the target objects in the protected agricultural area is realized,and the technical support for the development and management of the protected agriculture is provided.The main research work and results are as follows:(1)Through the acquisition and preprocessing of UAV image data such as clipping and segmentation,data enhancement,tag addition and so on,the sample image data of TOPA in UAV image can be obtained.The TOPA of the sample data is tagged according to the position and category of the image,and the sample dataset is divided into the training dataset and the test dataset to form the sample knowledge base.(2)A deep learning model based on Tensorflow is built and the TOPA recognition algorithm is designed.On the basis of analyzing the principle of CNN model and the characteristics of Tensorflow platform,the deep learning model is formed by designing and building the model framework.Then the TOPA recognition algorithm is designed based on the CNN model,and the image classification network is generated by using the improved convolution neural network structure and the proposal region extraction algorithm is designed on the basis of the region generation.The above methods are applied to the training and recognition of the sample dataset.(3)The TOPA recognition program is written to realize the automatic recognition and marking of TOPA on the UAV images,and the recognition results are analyzed and evaluated.Through the analysis of the experimental results,it is concluded that there are a few misunderstandings and leaks in the experiment affected by the interference factors such as light,cloud and so on.Secondly,the key factors affecting TOPA recognition are analyzed,and the recognition effect is optimized by adjusting the number of training steps,the number of convolution neural network layers and the number of nodes.Finally,the accuracy evaluation of the recognition results is given.The average recognition accuracy and success rate of TOPA are 92.12% and 83.27% respectively.Therefore,the recognition effect is better,which can provide a new method and reference for the recognition of similar ground objects in the high resolution remote sensing image.(4)Comparison and analysis of proposed method based on Tensorflow with different deep learning platform and traditional objects recognition method,it is concluded that the deep learning recognition method proposed in this paper is faster than the traditional ground object recognition method,and the recognition result is more intuitive.Moreover,compared with the five common methods in the deep learning framework of Caffe,the two platforms have the same accuracy and success rate,but the method proposed in this paper has more advantages in processing speed. |
