Deep Learning Model for Time-series Images to Discriminate Potato Crop in Punjab: Case Study of Monitoring Crop Harvesting

Today, with dynamically changing climatic conditions, it becomes very important to monitor and estimate the acreage and production of crops to meet the requirements of the bulging population. The deep learning models have been successfully applied in several domains and are being actively explored in the field of agriculture. This study explores the capability of deep learning models for specific crop mapping. The study was carried out in Ludhiana, Punjab, India for the level-3 classification of the potato crop. Time-series data was used to resolve the issue of overlapping spectral signatures of crops. The freely available Sentinel-2 and Sentinel-1 dataset was utilized in the study. Class-Based Sensor Independent Normalized Differential Vegetation (CBSI-NDVI) Index was used to enhance the target crop and separate it from the non-interest features in the study area. The optimum number of dates was selected for each target crop by performing separability analysis to generate the training data for the models. Two deep learning models were explored and optimized which were 1-Dimentional Convolution Neural Network (1D-CNN) model and an integrated Convolution Neural Network – Long ShortTermMemory(CNN-LSTM) model.AparametricModifiedPossibilistic c-Means (MPCM) classifier was tested to observe how it handled the heterogeneity within a class as compared to the deep learning models. Also, a dual-sensor approach was tested. The output generated was assessed using f1-score, precision and recall and using the Mean Membership Difference (MMD). The heterogeneity within a class was assessed using variance. It was observed that the integrated CNN-LSTM model generated best classification outputs and MPCM classifier produced results similar to the deep learning models.