DEEP CONVOLUTION NEURAL NETWORKS WITH RESNET ARCHITECTURE FOR SPECTRAL-SPATIAL CLASSIFICATION OF DRONE BORNE AND GROUND BASED HIGH RESOLUTION HYPERSPECTRAL IMAGERY

Abstract. Drones have been of vital importance in the fields surveillance, mapping, and infrastructure inspection. played a role acquiring high-resolution images with present need for precision farming, drones helped crop classification monitoring various patterns. With recent advancement computational power development robust algorithms to carry out deep feature learning neural network, based such techniques regained prominence contemporary research areas as common 2-D 3-D images, object detection, etc. In our research, we propose convolutional network architecture (CNN) aerial captured by Terrestrial Hyperspectral (THS or HSI) which includes 6-layers weights optimized along input layer, max-pooling fully connected softmax probability classifier, output layer. We acquired THS (using Cubert-GmbH data) drone agricultural data seasonal crops sowed during months March-June year 2017. Crop patterns include Cabbage, Eggplant, Tomato varying nitrogen concentrations region Bangalore, Southern India. To study influence impact CNN, ResNets model has applied. are combined followed recurrent (RCNN). The HSI layer corresponding ground truth is fed into spectral spatial residual 7*7*139 Imagery (HSI) volume. two blocks. An average pooling transform 5*5*24 spectral-spatial volume further single vector. At use an RMSProp optimizer error loss minimization when applied was able achieve overall accuracy 97.16%. Similarly, cabbage, eggplant tomato through same method achieved at 87.619%, 89.25%, 80.566% respectively comparison labels. ground-based datasets equipped good become promising tools improving quality efficiency agriculture today.