Information Extraction of Phyllostachys edulis Forests Distribution from UAV Images Based on Combination of Deep Learning and Object-Oriented Approach
[Objective]In order to provide technical reference for effectively estimating the area of bamboo forests,the paper tried to make information extraction of Phyllostachys edulis forests with a joint method of convolutional neural network-based (CNN )deep learning and object-oriented classification.[Method]Taking the Ph.edulis forests in research area as the research object and the UAV image as the data source,the deep learning model of CNN was used to identify the typical ground objects in the UAV image,and then the distribution information of Ph. edulis forests was extracted by using the object-oriented multi-scale segmentation and the classification method of the membership function.The accuracy of the classification results was compared with another accuracy extracted by using the random forest (RF)classification method.[Result](1 )The classification method adopted in this paper can not only avoid the difficulty of object-oriented classification method to make full use of the deep features of images,but also solve the problems of detailed information loss such as edge and shape of ground objects during image classification of CNN.(2)The overall accuracy of Ph. Edulis forests extracted by the classification method adopted in this paper was 91.04％,and the Kappa coefficient was 0.89;another overall accuracy of RF classification method was 85.07％,and the Kappa coefficient was 0.82.The overall accuracy was increased by 5.97 percentage points,Kappa coefficient was increased by 0.07.(3)According to the statistical data of forest,grassland and wet land resource results in 2023,the area of Ph.edulis forests in the study area was 150.31 hm2 .However,the area of Ph.edulis forests extracted with the classification method adopted in this paper was 161.99 hm2,and the extracted result was 11 .68 hm2 more than the actual result.In general,the degree of coincidence is high.[Conclusion]The joint method of deep learning based on CNN and object-oriented classification can improve the accuracy of information extraction,and it can also provide effective technical reference for the refinded extraction of Ph.edulis forests.
万方数据
维普
