The recognition and location of military scenes in hostile battlefield are of great strategic significance. Such scenes are the main targets of our long-range reconnaissance and directional strike. Deep transfer learning algorithm is always adopted to improve the accuracy of image recognition based on DCNN model. And on this basis, this paper mainly studied the application of deep transfer learning algorithm to recognize and locate typical scenes in large hyperspectral remote sensing image. Nichetargeting and impeccable DCNN model was accomplished after the training by typical scenes dataset. In the face of a large hyperspectral remote sensing image, the method of grid cutting, recognizing one by one and marking distinctively could pinpoint the location of typical scenes within. Experimental results showed that deep transfer learning algorithm could get a good application in the fast recognition and accurate location of typical scenes in large hyperspectral remote sensing image.
展开▼
