Evaluation of Deep Learning CNN Model for Land Use Land Cover Classification and Crop Identification Using Hyperspectral Remote Sensing Images

Deep learning convolutional neural network (CNN) is popular as being widely used for classification of unstructured data. Land use land cover (LULC) classification using remote sensing data can be used for crop identification also. Present study aims to examine the use of deep learning CNN for LULC classification on Indian Pines dataset and for crop identification on our study area dataset. In the present work, AVIRIS sensor's Indian Pines standard dataset has been used for LULC classification. Study area from Phulambri, Aurangabad, MH, India, has been used for crop classification. Data have been gathered from EO-1 Hyperion sensor. The accuracy of CNN model depends on optimizer, activation function, filter size, learning rate and batch size. Deep learning CNN is evaluated by changing these parameters. It has been observed that deep learning CNN using optimized combination of parameters has provided 97.58% accuracy for the Indian Pines dataset, while 79.43% accuracy for our study area dataset. The empirical results demonstrate that CNN works well in practice for unstructured data as well as for small size dataset.