Faculty Publications
Permanent URI for this communityhttps://idr.nitk.ac.in/handle/123456789/18736
Publications by NITK Faculty
Browse
4 results
Search Results
Item EXhype: A tool for mineral classification using hyperspectral data(Elsevier B.V., 2017) Adep, R.N.; Shetty, A.; Ramesh, H.Various supervised classification algorithms have been developed to classify earth surface features using hyperspectral data. Each algorithm is modelled based on different human expertises. However, the performance of conventional algorithms is not satisfactory to map especially the minerals in view of their typical spectral responses. This study introduces a new expert system named ‘EXhype (Expert system for hyperspectral data classification)’ to map minerals. The system incorporates human expertise at several stages of it's implementation: (i) to deal with intra-class variation; (ii) to identify absorption features; (iii) to discriminate spectra by considering absorption features, non-absorption features and by full spectra comparison; and (iv) finally takes a decision based on learning and by emphasizing most important features. It is developed using a knowledge base consisting of an Optimal Spectral Library, Segmented Upper Hull method, Spectral Angle Mapper (SAM) and Artificial Neural Network. The performance of the EXhype is compared with a traditional, most commonly used SAM algorithm using Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) data acquired over Cuprite, Nevada, USA. A virtual verification method is used to collect samples information for accuracy assessment. Further, a modified accuracy assessment method is used to get a real users accuracies in cases where only limited or desired classes are considered for classification. With the modified accuracy assessment method, SAM and EXhype yields an overall accuracy of 60.35% and 90.75% and the kappa coefficient of 0.51 and 0.89 respectively. It was also found that the virtual verification method allows to use most desired stratified random sampling method and eliminates all the difficulties associated with it. The experimental results show that EXhype is not only producing better accuracy compared to traditional SAM but, can also rightly classify the minerals. It is proficient in avoiding misclassification between target classes when applied on minerals. © 2016 International Society for Photogrammetry and Remote Sensing, Inc. (ISPRS)Item Prediction accuracy of soil organic carbon from ground based visible near-infrared reflectance spectroscopy(Springer, 2018) Minu, S.; Shetty, A.The present study was conducted to predict soil organic carbon (SOC) from ground visible near-infrared (Vis-NIR, 400- 2500 nm) spectroradiometer reflectance spectra. The objective was to study the effect of various pre-processing methods and prediction models on the accuracy of SOC estimated. Measured SOC content and reflectance spectra from pasture and cotton fields of Narrabri, Australia were used in the analysis. Reflectance spectra were pretreated with different smoothing methods such as: moving average, median filtering, gaussian smoothing and Savitzky Golay smoothing. A comparison between principal component regression, partial least square regression (PLSR) and artificial neural network models was carried out to get an optimum model for organic carbon prediction. The results indicate that PLSR model performs better with Savitzky Golay as the best pre-processing method for the study area, yielding R2cal = 0:84, RPDcal = 2.55 and RPIQcal = 4.02 in the calibration set and R2val = 0:77, RPDval = 2.17 and RPIQval = 3.19 in the validation set. The study recommends a suitable method in case of limited number of soil data. Based on the study, it can be said that properly pretreated reflectance spectra show tremendous potential in soil organic carbon prediction. © Indian Society of Remote Sensing 2017.Item Knowledge distillation: A novel approach for deep feature selection(Elsevier B.V., 2023) C, D.; Shetty, A.; Narasimhadhan, A.V.High dimensional data in hyperspectral remote sensing leads to computational, analytical, and storage complexities. Dimensionality reduction serves as an efficient tool to remove redundant, irrelevant, and highly correlated features. Recently, deep learning approaches have received remarkable progress in hyperspectral data analysis. In this paper, a new end-to-end deep learning framework based on a teacher-student network inspired by knowledge distillation is proposed for deep feature selection. Initially, a complicated teacher deep neural network is employed on complex high dimensional data to learn its corresponding best low dimensional representation. Then, the knowledge from the network is transferred to a simple student network that performs feature selection. Hence, it eventually leads to deep neural network compression which is of prime concern in hyperspectral remote sensing. Limited studies have been carried out to explore the benefits of knowledge distillation on hyperspectral data. The proposed method could be employed to choose deep features for both supervised and unsupervised tasks. Experimental results reveal the performance of the proposed scheme using limited features. In comparison to 1D and simple autoencoder models, the 2D model based on convolutional autoencoder delivers greater classification accuracies, with a classification accuracy value of 96.15% for the Indian Pines dataset and 97.82% for the Pavia University dataset. A similar trend is reported with unsupervised learning as well. Furthermore, the proposed model has a low degree of sensitivity to parameter selection. © 2022 National Authority of Remote Sensing & Space ScienceItem MICAnet: A Deep Convolutional Neural Network for mineral identification on Martian surface(Elsevier B.V., 2024) Kumari, P.; Soor, S.; Shetty, A.; Koolagudi, S.G.Mineral identification plays a vital role in understanding the diversity and past habitability of the Martian surface. Mineral mapping by the traditional manual method is time-consuming and the unavailability of ground truth data limited the research on building supervised learning models. To address this issue an augmentation process is already proposed in the literature that generates training data replicating the spectra in the MICA (Minerals Identified in CRISM Analysis) spectral library while preserving absorption signatures and introducing variability. This study introduces MICAnet, a specialized Deep Convolutional Neural Network (DCNN) architecture for mineral identification using the CRISM (Compact Reconnaissance Imaging Spectrometer for Mars) hyperspectral data. MICAnet is inspired by the Inception-v3 and InceptionResNet-v1 architectures, but it is tailored with 1-dimensional convolutions for processing the spectra at the pixel level of a hyperspectral image. To the best of the authors’ knowledge, this is the first DCNN architecture solely dedicated to mineral identification on the Martian surface. The model is evaluated by its matching with a TRDR (Targeted Reduced Data Record) dataset obtained using a hierarchical Bayesian model. The results demonstrate an impressive f-score of at least .77 among different mineral groups in the MICA library, which is on par with or better than the unsupervised models previously applied to this objective. © 2024