Agricultural Crop Hyperspectral Image Classification using Transfer Learning

In recent years, there is increasing interest around developing efficient Deep learning methods using convolutional neural networks (CNNs) in classifying Hyperspectral images (HSI). The performance of these networks highly depends on the availability of ample amount of labelled samples for training. To solve the problem of insufficient training samples, Transfer Learning (TL) is currently being incorporated in deep networks. The main objective and purpose of this paper is to implement a model that performs classification task quickly with high performance and is also both resource and data efficient. VGGNet and ResNet networks trained on benchmark ImageNet dataset are considered as source models and learned features from these networks, are transferred to new model that is to be trained on target data. The proposed model is tested on two popular datasets (i.e., Indian pines and Salinas) along with a novel dataset containing field crop data of Kota region in Rajasthan. Experimental results demonstrate that TL based model can achieve remarkable accuracy even with small-training samples on target data.