This research work aims to develop a deep learning-based crop classification framework for remotely sensed time series data. Tobacco is a major revenue generating crop of Khyber Pakhtunkhwa (KP) province of Pakistan, with over 90% of the country's Tobacco production. In order to analyze the performance of the developed classification framework, a pilot sub-region named Yar Hussain is selected for experimentation work. Yar Hussain is a tehsil of district Swabi, within KP province of Pakistan, having highest contribution to the gross production of the KP Tobacco crop. KP generally consists of a diverse crop land with different varieties of vegetation, having similar phenology which makes crop classification a challenging task. In this study, a temporal convolutional neural network (TempCNNs) model is implemented for crop classification, while considering remotely sensed imagery of the selected pilot region with specific focus on the Tobacco crop. In order to improve the performance of the proposed classification framework, instead of using the prevailing concept of utilizing a single satellite imagery, both Sentinel-2 and Planet-Scope imageries are stacked together to assist in providing more diverse features to the proposed classification framework. Furthermore, instead of using a single date satellite imagery, multiple satellite imageries with respect to the phenological cycle of Tobacco crop are temporally stacked together which resulted in a higher temporal resolution of the employed satellite imagery. The developed framework is trained using the ground truth data. The final output is obtained as an outcome of the SoftMax function of the developed model in the form of probabilistic values, for the classification of the selected classes. The proposed deep learning-based crop classification framework, while utilizing multi-satellite temporally stacked imagery resulted in an overall classification accuracy of 98.15%. Furthermore, as the developed classification framework evolved with specific focus on Tobacco crop, it resulted in best Tobacco crop classification accuracy of 99%.

Due to the rapid growth of population the food need also increases which is the center of focus for various researchers and governments. For this purpose crop information system has been made, the aim of crop information system is to monitor the crop health and estimate the needs for the next four to five years. Geo graphic information system plays an important role in crop estimation and identification. GIS uses remote sensing technique to identify various crops and their yield. In this paper novel approaches are used for the identification and estimation of tobacco. SPOT 5 imagery having resolution of 2.5m is used for the estimation and identification of tobacco. For post processing, statistics like kappa coefficients and Receiver operating curves are utilized. This study mainly focuses on the Mardan region in KPK Pakistan. Classification is done for four categories, these categories are then classified using state of the art machine learning classifiers and the accuracy of these various classifiers has been compared.

Deep learning based data driven methods with multi-sensors spectro-temporal data are widely used for pattern identification and land-cover classification in remote sensing domain. However, adjusting the right tuning for the deep learning models is extremely important as different parameter setting can alter the performance of the model. In our research work, we have evaluated the performance of Convolutional Long Short-Term Memory (ConvLSTM) and deep learning techniques, over various hyper-parameters setting for an imbalanced dataset and the one with highest performance is utilized for land-cover classification. The parameters that are considered for experimentation are; Batch size, Number of Layers in ConvLSTM model, and No of filters in each layer of the ConvLSTM are the parameters that will be considered for our experimentation. Experiments also have been conducted on LSTM model for comparison using the same hyper-parameters. It has been found that the two layered ConvLSTM model having 16-filters and a batch size of 128 outperforms other setting scenarios, with an overall validation accuracy of 97.71%. The accuracy achieved for the LSTM is 93.9% for training and 92.7% for testing.

No abstract