Understanding the Issues involved in Crop Classification Using Microwave and Optical Remote Sensing Data

Neetu

Please use this identifier to cite or link to this item: http://10.1.7.192:80/jspui/handle/123456789/10502

Full metadata record

DC Field	Value	Language
dc.contributor.author	Neetu	-
dc.date.accessioned	2022-01-21T06:12:36Z	-
dc.date.available	2022-01-21T06:12:36Z	-
dc.date.issued	2020-10	-
dc.identifier.uri	http://10.1.7.192:80/jspui/handle/123456789/10502	-
dc.description	ST000078	en_US
dc.description.abstract	Agriculture is a vital sector of the Indian economy. In India, many crops are grown including cereals, millets, pulses, oilseeds, commercial crops, and horticultural crops. Directorate of Economics and Statistics (DES), under the Ministry of Agriculture and Farmer's Welfare of India, provides advance estimates for the country's major crops. The labor-intensive conventional methods provide these multiple estimates for crop area and production through different sampling techniques. However, the conventional estimates are not digital, and these are time-consuming and subject to human errors. Therefore, for timely reporting and to avoid human bias, remote sensing (RS) technology was introduced for crop assessment in the 1980s. Success of the RS technology is dependent on the accurate analysis of crop area satellite images. Therefore, the crop classification within a satellite image/data is one of the most significant components of crop area estimation using RS technology. However, the limited availability of cloud-free data, small field size, actual ground level crop data, and robust methodology to generate accurate crop maps at high resolution for Kharif crops such as pigeon pea, cotton, soybean, etc., are some of the challenges for crop area estimation in India. The crop area mapping at different geographical or administrative levels such as district/block/farm levels is another significant challenge for the wide spread use of RS technology. The possible solution to these challenges may be advanced classification techniques such as machine learning and artificial intelligence along with the use of both optical and microwave satellite data. In this context, this Ph.D. research work was carried out with three objectives, i.e., i) Comparative evaluation of various classifiers for better crop classification, ii) Evaluating the suitability of multi-polarization SAR data for crop discrimination, and iii) Assessing optimum combination (Multi-date, multi-resolution, and SAR + Optical) of satellite data for crop classification. To achieve these objectives, the best available high-resolution remote sensing data from various satellites/sensors (optical and microwave) with a significant amount of ground truth was analyzed with various available online geospatial tools GEE and SNAP. iv Institute of Science, NIRMA University, Ahmedabad Under the first objective, crop classification efficiency of different machine learning techniques such as CART, RF, SVM with MLC using optical data (Sentinel-2) was evaluated in the research farm of IARI. The accuracy assessment with user's and producer's accuracy, kappa coefficient, and F measure was carried out, and results were validated with the actual farm data. It was possible to classify wheat and mustard using all four classification techniques. Further, the vegetable crop classification had low accuracy, 33.5% using MLC, but improved to 65.7% using SVM and 91.9% using the RF algorithm. Under the second objective, two case studies were carried out using multi-date and multi-parametric SAR data for crop discrimination. In the first case study, multi-date and multi-polarization SAR from RISAT-1 Fine Resolution Strip map Mode-2 (FRS- 2) quad polarization was used to study the separability of various crops (wheat, mustard, gram, and pigeon pea). The analysis of multi-date single-polarization and single-date multi-polarization were done for various studied crops. A clear separation of mean σ0 values of various land cover classes was observed. However, standard deviations of different crop classes were overlapping. With multi-date SAR, better discrimination was achieved between various crop classes using backscatter values' temporal profile. In the second study, multi-parametric and multi-date SAR data from Sentinel-1 (VV polarization) and Radarsat-2 (HH, VV, VH, and HV polarization) were used for crop discrimination. The best satellite dates/layers were selected for analysis using a comprehensive approach of combining PCA, separability analysis, and correlation analysis. Machine learning techniques such as RF, SVM, SAM, and MLC were evaluated for crop classification (soybean, rice, and other crops). Additionally, Radarsat-2 fully polarimetry data was explored to classify rice and soybean crops using Freeman Durden decomposition and Wishart classification. The accuracy assessment was implemented with ground observations and found that the soybean crop had the highest classification accuracy (F measure 70.3% to 87.9%), and SVM outperformed other classification techniques. Under the third objective, two case studies were conducted. In the first study, multidate single and dual-polarization SAR (RISAT-1) and Resourcesat-2 AWiFS data were used to classify rice and cotton crops in the Sirsa district of Haryana. The crop area estimated using multi-date RISAT HH data was compared with government estimates PH. D. Thesis (Neetu, 2020) Institute of Science, NIRMA University, Ahmedabad v and the crop area statistics generated under the FASAL project of Government of India. Though the estimated rice area matched very well (-1.6% deviation) with the FASAL estimates, there was an underestimation (-14.4%) for the cotton area. In the second case study, data fusion of SAR and the optical images were attempted at the pixel level (using Brovey, PCA, Multiplicative & Wavelet with Intensity Hue and Saturation (IHS), feature level (adding NDVI as an additional feature to 3-date SAR images), and at the decision level (using outputs derived from SAR data and then overlaying on optical data) for better crop discrimination. The study was conducted for the Yadgir district of Karnataka. The feature-level fusion approach provided the maximum accuracy for rice crops, while for cotton and pigeon pea crops, decision-level fusion improved the accuracy from less than 60% to over 84%. Thus, in this Ph.D. research work, different advanced classification techniques were attempted on various types of satellite data for crop classification in different agro ecological and administrative regions of the country. The results showed the usefulness of machine learning algorithms, polarimetric SAR data, and combined optical and microwave SAR data. However, there is a scope to study further to develop a technique for classification of more Kharif crops such as sorghum, maize, groundnut, castor, pulses, etc. These crops estimation are required for policy-making and export-import decisions. With the limitations and advantages of satellite data, further research may be carried out using both optical and microwave data for better crop classification	en_US
dc.description.sponsorship	Guided by Dr. Shibendu S. Ray	en_US
dc.language.iso	en_US	en_US
dc.publisher	Institute of Science, Nirma University	en_US
dc.relation.ispartofseries	;ST000078	-
dc.subject	Science Theses	en_US
dc.subject	Theses 2020	en_US
dc.subject	14EXTPHDS60	en_US
dc.subject	Microwave and Optical Remote Sensing Data	en_US
dc.title	Understanding the Issues involved in Crop Classification Using Microwave and Optical Remote Sensing Data	en_US
dc.type	Thesis	en_US
Appears in Collections:	Theses, IS

Files in This Item:

File	Description	Size	Format
ST000078.pdf	ST000078	8.42 MB	Adobe PDF	View/Open

Show simple item record

IR @ Nirma University