Research On Jujube Field Classification And Recognition Based On High Resolution Remote Sensing Data

As a large-scale planting characteristic forest and fruit advantage economic crops in Xinjiang,jujube plays an increasingly important role in Xinjiang oasis production and ecological construction because of its wide planting range and large planting area,so quickly and accurately understand the actual situation of jujube planting area and planting structure,not only from the local area to serve the production operation management,but also to the relevant department scientific decision-making and management are of great significance.With the development of domestic high-resolution satellite remote sensing technology,high resolution series satellites are applied in the field of agricultural remote sensing with the advantages of high spatial resolution and obvious detailed information of surface features,which can provide more sophisticated agricultural services.Focusing on an accurate detection of large-scale jujube orchards/fields in the southern Xinjiang Uygur Autonomous region with high efficiency to satisfy the needs of farming and agricultural management and decision-making,based on the object-oriented classification technology and deep learning technology,this paper adapts the preprocessed fusion gf-6 image data source,first studies 224 Regiment(Kunyu city)of 14th division of Xinjiang Production and Construction Corps,and then analyzes the application of two methods in the application test area,so as to study the feasibility of object detection and recognition in jujube orchard based on generalized transfer deep learning.The main work and achievements are as follows.(1)Based on object-oriented classification technology,the optimal scale segmentation algorithm is explored to select the optimal segmentation parameters,including heterogeneity factor,scale parameter information,band weight and so on.Combined with the auxiliary data of edge detection operator in order to avoid the over-segmentation and under-segmentation in the process of multi-scale segmentation,which can improve the image segmentation quality and obtain more accurate large-scale jujube orchard information.According to the accuracy evaluation,the overall accuracy and Kappa coefficient of the object-oriented classification of jujube are 0.77 and 0.68,and the user accuracy and producer accuracy of jujube gardens are 0.92 and 0.82.(2)Based on deep learning technology,the current popular target detection algorithm(Faster R-CNN algorithm)is studied and applied to the regional jujube field recognition in this paper.In order to improve the recognition and detection effect of jujube field,the corresponding jujube data set and the data generalization expansion method are made,and the vgg16 network model pre-trained on large data set ImageNet is used for migration learning to improve the generalization ability of the model and determine the best network model.The average recognition precision based on deep learning model algorithm is 0.979,the recall rate is 0.952,and F1-score is 0.965 by verification set.(3)Based on different algorithms such as object-oriented classification technology and deep learning technology to further study the affection of jujube field information in the application area,the results show that the accuracy of object-oriented classification is 0.82,kappa coefficient is 0.62.According to the statistics of deep learning model method,the overall classification accuracy is 0.97 and kappa coefficient is 0.93,which is 0.12 and 0.31 higher than the former method,respectively.It is known that the deep learning algorithm can be effectively meet the accuracy requirements,and the operability of the experimental method(Faster R-CNN algorithm)has been verified.