Browsing by Author "Phartiyal, Gopal Singh"

Now showing 1 - 20 of 21

Comparative study on deep neural network models for crop classification using time series polsar and optical data
(ISPRS, 2018-11) Phartiyal, Gopal Singh
Crop classification is an important task in many crop monitoring applications. Satellite remote sensing has provided easy, reliable, and fast approaches to crop classification task. In this study, a comparative analysis is made on the performances of various deep neural network (DNN) models for crop classification task using polarimetric synthetic aperture radar (PolSAR) and optical satellite data. For PolSAR data, Sentinel 1 dual pol SAR data is used. Sentinel 2 multispectral data is used as optical data. Five land cover classes including two crop classes of the season are taken. Time series data over the period of one crop cycle is used. Training and testing samples are measured and collected directly from the ground over the study region. Various convolutional neural network (CNN) and long short-term memory (LSTM) models are implemented, analysed, evaluated, and compared. Models are evaluated on the basis of classification accuracy and generalization performance.
Computational-vision based orthorectification and georefrencing for correct localization of railway track in UAV imagery
(IEEE, 2021-02) Phartiyal, Gopal Singh
In recent years, reliable rail track health monitoring and localization requires accurately orthorectified and georeferenced imagery. The vision-based approach is most suited for the geometrical correction of unmanned aerial vehicles (UAVs) high-resolution imagery in the given scenario. The single image acquired by UAV covers a significant area and contains only one reference point and many distorted pixels. This paper provides a novel computational vision-based approach for orthorectification and georeferencing of a single rail track aerial image among the set of given images without an exclusive reference map of that location and ground control points.
A critical analysis of polarimetrie signatures on PALSAR 2 data for land cover classification
(IEEE, 2016) Phartiyal, Gopal Singh
In this paper, polarization signatures are extracted for utilization of the fully polarimetrie L-band ALOS-PALSAR 2 data. These signatures are extracted for different land cover classes (i.e., urban, water, short vegetation, tall vegetation and bare soil). Critical analysis is performed on the polarization signatures generated of different classes. Further, polarization signatures of different classes are compared with the help of normalized Euclidean distance (NED) and normalized signature correlation mapper (NSCM). Decision tree based algorithm is developed with the help of NSCM, NED and backscattered image for the classification of different land cover classes.
Deep-domain adaptation approach based crop disease prediction using UAV photometry images
(Science Nexus, 2023) Phartiyal, Gopal Singh
Agriculture is the backbone of any community, as it provides the most necessity for human survival. Diseases on plants/crops in agriculture reduces the productivity, thus its presence and removal are mandatory for good yield. Some of the common symptoms of disease in plants disease are leaf rust, stem rust, powdery mildew, leaf spot, birds-eye spot on berries, damping off of seedlings, sclerotinia, and chlorosis. These diseases can be identified visually by observing the physical condition of a plant’s leaves.This paper proposes the domain adaptation based model to predict the diseases in the crop in their early ages using UAV imagery. This gives information related to the water system, soil variety, pests, and fungal infestations. Crops images, collected by the UAVs, have information in the range of optical, especially visual spectra. Different features from these images can be extracted, which gives information about the health of plants in a manner that cannot be overserved otherwise. Another important feature of UAV based monitoring is its techno-economical approach to monitor the growth consistently and regularly i.e. at each month, day, or hourly basis if needed. The availability of crop information at this frequency helps scientists and farmers to take timely conter-measure decisions and actions. The proposed model has been developed using EfficientNet. The proposed model trained and tested on PlantVillage dataset. This dataset includes over 87K RGB images of healthy and diseased crop leaves, divided into 38 distinct classes. To overcome with the class imbalance problem, image enhancement, augmentation & rotation has been adopted. The complete dataset is splitted into 80/20 ratio of training and testing sets while maintaining the directory structure. Afterwards, model is trained and fine-tuned with UAV Photometry images using domain adaptation transfer learning approach.This leads to risk minimization.The model utilizes the methodologies of domain-invariant spaces and feature augmentation. The performance of the model evaluated in terms of specificity, sensitivity, accuracy and precession resulted in satisfactory performance with accuracy leading up to the order of 90%. Further, the proposed models are of lightweight in nature, which expands applicability and flexibility of model.
Denoising of fundus images using feed-forward convolutional neural networks
(IEEE, 2024-01) Phartiyal, Gopal Singh
The biomedical image denoising method has developed into one of the most fascinating study fields today. Every day, lots of biomedical images are taken, and it is from these images that diseases have been diagnosed. To diagnose eye-related diseases, a fundus image is taken. Early detection of eye-related illnesses is essential to prevent severe problems like cataracts and blindness in the future. The analytical procedure is hampered by a noisy fundus, which makes diagnosis difficult. As a result, it's necessary to lessen noise without sacrificing image quality (blood vessels, optic disk, macula, hemorrhage, and exudate). Therefore, classical denoising methods, transform denoising technique and Deep Learning based denoising method have been utilized to diminish these noises and results have been compared based on structural similarity Index matrix (SSIM) and peak signal to noise ratio (PSNR). Finally, Feed-forward Denoising Convolutional Neural Networks (DnCNN) technique outperforms over all others conventional as well as Deep Learning (SDCDAE, GAN-CT, RED-CNN, and CNN-DWT) denoising methods. In DnCNN Residual learning and Batch normalization (RL and BN) have been utilized to accelerate task of training in addition boosts the denoising performance on the same time, preservers the important features of images. DnCNN technique has been provided better results on the specified noise levels as well as unknown noise levels.
Design of an I-band microstrip patch antenna using dual-superstrate layers for microwave imaging system
(IEEE, 2023-07) Phartiyal, Gopal Singh
The development for a microwave imaging system, better resolution and good penetration depth are the most prominent constraints. To achieve these requirements, high gain antenna operating at low frequency ranges with considerable bandwidth is needed. As, wider bandwidth improves the range resolution and high gain improves the range. Therefore, the L-band antenna is a suitable choice for further investigation and research. While designing the antenna, it is observed that antenna size is one of the important constraints for the development of imaging system. Therefore, this research article examines the effect of gain improvement for a dielectric superstrate and slot in the microstrip patch antenna that can be efficiently used for microwave imaging applications. The proposed design of antenna consists of simple geometry. However, lower gain, low radiation efficiency, and narrow bandwidth are the key digression factors associated with microstrip patch antenna. To overcome these limitations, the use of superstrate with slot in patch antenna are studied here. A dual layer superstrate approach is utilized to achieve improved gain along with high F/B ratio in this paper. The development stages of the proposed antenna are discussed and evaluated. Furthermore, parametric analysis is also conducted and reported for better observations.
Fractional crop cover estimation via drone imagery and machine learning with color models
(IEEE, 2024) Phartiyal, Gopal Singh
Drones have become increasingly popular in precision agriculture due to their ability to collect valuable data quickly and efficiently. One of the major aspects of precision agriculture is to estimate fraction crop cover at an early stage. This paper develops an approach using fine-tuned Machine Learning (ML) YOLO models to extract crop fields only from drone imagery and mask all other objects. A combination of Color Space Models (CSM) is used to extract fraction crop cover at an early stage. The approach was developed for extracting information with less processing complexity as drones/UAVs have limited processing and power capabilities. The primary objective of this study is to identify low-density crop areas and barren land within crop fields during the early stages of crop growth using CSM. Otsu and Max Entropy thresholding techniques are analysed to obtain mask information of targeted area. Morphological open and close operations are used to get desired size patches of sparse or no crop location. The study suggests Otsu thresholding as an adaptive thresholding method as its results are adequate compared to ground truth. Different filter size results are also compared as filter size determines the minimum patch size identified on the ground. The finetuned ML model extracts the object of interest. The color space model works well when applied to that single object.
Impact of permuted spectral neighborhood of high-dimensional msts rs data on crop classification performance with DNN models
(IEEE, 2023-10) Phartiyal, Gopal Singh
It is still a challenge for existing DNN based models to synergistically exploit the spatial, temporal, and especially spectral information of a crop present in multi-sensor time series (MSTS) remote sensing (RS) images and provide accurate crop classification while keeping the generalization ability of DNN models high. This imbalance requires investigation and demands novel CNN and RNN model-based approaches that can address the issue. The novel models proposed in this study involve the concepts of permuted localized spectral convolutions, localized spatial convolutions, and bi-directional recurrent units. The permuted spectral band stacking strategy is explored in this study to strengthen the influence of the spectral information. Overall, 6 models are proposed namely; Perm-1D-CNN, Perm-3D-CNN, Perm-RNN, Perm-1D-CRNN, Perm-2D-CRNN, and Perm-3D-CRNN. The qualitative and quantitative assessments reflect the higher generalization ability of the Perm-3D-CRNN along with its high classification accuracy. Also, the impact of spectral band permutations and localized spectral convolutions on the performance of DNN models is significant toward improved generalization.
An improved land cover classification using polarization signatures for PALSAR 2 data
(Elsevier, 2020-06) Phartiyal, Gopal Singh
Land cover classification in mixed land cover scenarios is challenging with PolSAR data. Polarimetric decomposition techniques are most popular methods for PolSAR data classification in recent times. These techniques focus on identification of dominant scattering phenomena and hence result in sub-optimal classification in mixed land cover scenarios. Alternatively, polarization signatures (PSs) are good illustrations of SAR target responses as they depict a detailed physical information from target backscatter. Researchers have successfully utilized SAR PSs for land cover (LC) classification. Some reports suggested utilizing correlation between observed PSs and standard target PSs as features for LC classification. This paper presents a study on improved utilization of PSs for optimal LC classification in mixed class scenarios. First, PS based SAR features are derived using fully polarimetric SAR data. The features represent a degree of similarity between observed and standard PSs. The derived features are termed as polarization signatures correlation features or PSCFs. The novel PSCFs are analyzed, evaluated and compared with decomposition based features for the purpose of LC classification. Classification performance indicators highlight potential of PSCFs for mixed LC classification problems. Therefore, further an adaptive and optimal LC class boundary estimation approach for LC classification is proposed and developed. Observed PSs and reference LC class PS statistics are used to build empirical models between classification performance indicators and LC class boundaries. The empirical models are optimized using the evolutionary genetic algorithm to maximize classification performance. A decision tree is constructed based on the optimal class boundaries to prepare LC classification. The proposed classification approach is compared with some recent popular classifiers and comparison suggests that the proposed approach provides satisfactory results for mixed LC classification scenarios.
Improved mapping of flood affected villages in India: a novel three-stage approach using PolSAR polarization signatures and ensembled dilated CNNs
(Taylor & Francis, 2023-11) Phartiyal, Gopal Singh
During floods, updated and accurate information on affected human settlements helps save lives and reduces time to rescue. Therefore, approaches that can provide reliable information during floods via the use of all-weather and real-time functional technology is highly needful. The study presented here attempts to efficiently and precisely map villages in the Indian sub-continent during floods via a three-stage approach which uses PolSAR data. However, an accurate segregation of villages in India even with PolSAR data is challenging because the built-up structures in the villages of rural India are closely placed and are randomly oriented w.r.t. each other. This condition either hinders their segregation or otherwise induces false alarms during extraction. More descriptive land cover characterization features and powerful feature classifiers may address this challenge. The study in this paper proposes a novel approach to efficiently detect and map flood affected villages which utilize polarization signatures from PolSAR imagery, ensemble-of -dilated-convolutions based CNNs, apriori knowledge and image morphology. The approach broadly involves three stages: first, built-up area extraction from a PolSAR image: second, detection of villages in a built-up area image and third, identification and mapping of villages that are affected by the flood. In the first stage, an ensemble of varying dilated-convolutions based novel CNN classifier which directly utilizes PolSAR-2 polarization signatures (PSs) in window-mode as features are developed to extract built-up areas. The second stage provides a novel village detection filter based on apriori knowledge and image morphology to detect actual villages and mask out the false objects. Finally, in the third stage, flood affected villages are mapped via a series of morphological operations based degree-of-intersection measure. Experiments are conducted on both simulated and natural flooded area datasets. Experimental results show 81% detection accuracy and 100% mapping performance of the proposed approach which indicates its potential as an effective flood affected village mapping system.
Improved utilization of polsar polarization signatures using convolutional-deep neural nets for land cover classification
(IEEE, 2019-11) Phartiyal, Gopal Singh
Normalized Euclidean distance (NED) and normalized signature correlation mapper (NSCM) are most popularly used pattern classifiers with polarization signatures (PSs) based polarimetric synthetic aperture radar (PolSAR) data applications. These methods are not able to fully exploit the PSs as they do not exploit the spatial context or pattern of PSs which is essential. Improved utilization of PSs is still required for PolSAR applications such as agriculture crop classification and monitoring. In this study, convolutional deep neural networks (C-DNNs) are introduced and utilized as pattern classifiers for PS classification. C-DNNs have the ability to consider and control the influence of local neighborhood pixels during classification. Therefore, in this study C-DNNs are utilized to extract and exploit subtle changes between PSs of land covers to improve classification performance. Comparison with NED and NSCM classifiers signify the contribution of C-DNNs by improved performance in PolSAR data classification.
Introducing ISAP and MATSS: mental stress induced speech utterance procedure and obtained dataset
(Elsevier, 2022-11) Phartiyal, Gopal Singh
Mental stress persisting for long can cause severe health issues. There are various approaches available in the literature for investigating stress through speech utterances. The available procedure to obtain speech under stress dataset requires the speakers to undergo the actual stress situations in a real environment with limited control or inducing stress with a mental task in a lab environment. These approaches either suffer from ethical issues or unreliable labeling of the obtained speech samples. In this paper, we attempt to overcome these limitations with Induced mental Stress based speech production And labeling Procedure (ISAP), for obtaining speech utterances under mental stress along with labeling the samples simultaneously. The proposed ISAP can be incorporated by future studies as per their need to create a speech under stress dataset. We also present the obtained dataset, the baseline experiments, and classification results with various machine learning models. A total of 1260 speech utterances are obtained, with ISAP able to induce stress in 54.4% of the cases. The accuracy of the SVM classifier in recognizing three stress classes, namely, No Stress, Low Stress, and High Stress is found to be 57.1%.
Land cover mapping of mixed classes using 2D CNN with multi-frequency SAR data
(Elsevier, 2024-07) Phartiyal, Gopal Singh
Synthetic aperture radar (SAR) data obtained at multiple frequencies and polarizations offers valuable complementary information for classifying mixed classes that exhibit similar backscattering response. Although deep learning-based convolutional neural networks (CNNs) effectively extract features from multi-frequency SAR data, the arbitrary ordering of SAR features may hinder optimal convolution of the best feature sub-space for a specific class and underutilize available multi-frequency data. To address this, a novel CNN transforming SAR feature-space from 1-D to 2-D and employing varied dilation-rate convolutions is introduced. This transformation maximizes unique and localized feature combinations, efficiently utilizing the available feature sub-spaces and extracting discriminative features for accurate classifications, addressing the challenge of arbitrary band neighborhoods. Utilizing dual-polarization SAR data from ALOS-2 PALSAR-2 and Sentinel-1 sensors, the proposed CNN achieves an average f-score of 0.97 and a kappa coefficient of 0.97, an improvement of 11 %, 7 % and 3 % in OA compared to the 1-D, 2-D and 3-D CNN classifiers, without feature transformation. The classifier's generalization ability is evaluated using ground truth knowledge of various heterogeneous classes, and the proposed CNN classifier outperforms others in terms of accuracy metrics and generalization ability.
A mixed spectral and spatial convolutional neural network for land cover classification using SAR and optical data
(EGU 2018, 2018) Phartiyal, Gopal Singh
Today, both SAR and optical data are available with good spatial and temporal resolutions. The two data modalities complement each other in many applications. There are numerous approaches to process the two data modalities, separately or combined. Domain or modality specific approaches such as polarimetric decomposition techniques or reflectance based techniques cannot work with the two datasets combined together. Data fusion approaches incur information loss during the process and are highly application specific. Machine learning (ML) approaches can operate on the combined dataset but have their own advantages and disadvantages. There is a need to explore new ML based approaches to achieve higher performance. Convolutional neural networks (CNNs) are young, trending, and promising ML tools in remote sensing applications. CNNs have the capability to learn complex features exclusively from data. Data from the two modalities can thus be brought together and processed with increased performance. In this paper an attempt is made to analyze CNN capabilities to perform land cover classification using multi-sensor data. SAR data used in this study is L band fully polarimetric PALSAR 2 data with 6 meter spatial resolution. Three basic polarimetric bands, namely, HH, HV, and VV, and four derived bands (polarization signatures) are used. Six multispectral Landsat 8 bands, pan sharpened and resampled at 6 meter spatial resolution, are used as optical data. All 13 features are stacked together and fed as input data to the proposed CNN. The areas selected for study are Haridwar and Roorkee regions of northern India. This study introduces a CNN where convolution is performed both spatially and spectrally. We show how this is an advantage over performing only spatial convolution. Five land cover classes namely, urban, bare soil, water, dense vegetation, and agriculture are considered. The CNN is trained on more than 1200 ground truth class data points measured directly on the terrain. The classification shows results with good generalization. Comparison with other classifiers such as SVMs shows that the proposed approach provides better classification results in terms of generalization, although the cross-validation accuracy is on the same order. The evaluation of the generalization of the classified image is done using ground truth knowledge on selected subset areas in the study area.
Modeling and analysis of foliage environment using wideband radar system
(IEEE, 2023-12) Phartiyal, Gopal Singh
Foliage penetrating (FOPEN) radar can see through foliage in clear weather as well as in presence of dust, smoke, rain and haze. Foliage is a complex clutter environment and target detection in this environment is a highly critical task. The dielectric constant of foliage depends upon the bulk density and gravimetric moisture content of foliage. These parameters depend upon the crop growth stages and crop density. This paper presents an approach for modelling and analysis of foliage environment backscatter, for critical understanding of the backscattering behaviour of targets in complex clutter environment. Foliage environment is modelled and A- and, B- scans are analyzed having two different dielectric contrast targets (Aluminum and Teflon) using sugarcane as foliage. The proposed modelling is helpful to develop an experimental wide band FOPEN system for dense foliage system to achieve fine range resolution.
Non iterative LDPC decoding by syndrome generation using artificial neural network
(IEEE, 2016-07) Phartiyal, Gopal Singh
Low density parity-check code (LDPC) is an error correcting code used in noisy communication channel (e.g. AWGN) to reduce the probability of error in information. By using LDPC codes, this probability can be made comparatively small, so that the data transmission rate can be as close to Shannon's limit. The decoding of Low Density Parity Check (LDPC) codes by iterative process of belief propagation gives challenges for designers looking for real time performance in communication systems. This thesis work proposes the use of Artificial Neural Networks (ANN) to replace belief propagation to approach closer to Shannon's limit more closer than other traditional decoding methods. This thesis is intended to design a new methodology to decode LDPC codes in Non-iterative manner with the help of ANN and Look Up Table (LUT). This work is at initial stage and will be extended for better performance.
Optimal use of polarimetric signature on PALSAR-2 data for land cover classification
(IEEE, 217) Phartiyal, Gopal Singh
SAR data is playing key role in monitoring, the current status or change in, the land cover. For unsupervised SAR image classification, polarization signatures can play a significant role. Since it is difficult to obtain specific polarization signature of real land cover, it is customary to represent them with standard canonical structures polarization signatures. A critical analysis of the complex signatures of real targets is essential thereafter it is also a challenge to decide the thresholds or class boundary value on the correlation images. Therefore, in this paper an attempt has been made to critically analyze the polarimetric signature of complex targets and based on the correlation image analysis an OTSU multi-thresholding based approach is proposed to decide the individual class boundary values which will finally help in building a decision tree (DT) based classification technique. For this purpose L band fully polarimetric SAR data (PALSAR-2) has been used. DT class thresholds are computed using OTSU multi-thresholding method, scatter plot method, and a priori information. Obtained results reveal that complementary features like polarization signatures can help in identification as well as classification of land surface objects significantly by the proposed method.
Permuted spectral and permuted spectral-spatial cnn models for polsar-multispectral data based land cover classification
(Taylor & Francis, 2020-12) Phartiyal, Gopal Singh
It is a challenge to develop methods which can process the polarimetric synthetic aperture radar (PolSAR) and multispectral (MS) data modalities together without losing information from either for remote sensing applications. This paper presents a study which attempts to introduce novel deep learning-based remote sensing data processing frameworks that utilize convolutional neural networks (CNNs) in both spatial and spectral domains to perform land cover (LC) classification with PolSAR-MS data. Also since earth observation remotely sensed data have usually larger spectral depth than normal camera image data, exploiting the spectral information in remote sensing (RS) data is crucial as well. In fact, convolutions in the sub-spectral space are intuitive and alternative to the process of feature selection. Recently, researchers have gained success in exploiting the spectral information of RS data, especially the hyperspectral data with CNNs. In this paper, exploitation of the spectral information in the PolSAR-MS data via a permuted localized spectral convolution along with localized spatial convolution is proposed. Further, the study in this paper also establishes the significance of performing permuted localized spectral convolutions over non-localized or localized spectral convolutions. Two models are proposed, namely a permuted local spectral convolutional network (Perm-LS-CNN) and a permuted local spectral-spatial convolutional network (Perm-LSS-CNN). These models are trained on ground truth class data points measured directly on the terrain. The evaluation of the generalization performance is done using ground truth knowledge on selected well-known regions in the study areas. Comparison with other popular machine learning classifiers shows that the Perm-LSS-CNN model provides better classification results in terms of both accuracy and generalization.
Synergistic exploitation of localized spectral-spatial and temporal information with DNNs for multisensor-multitemporal image-based crop classification
(Elsevier, 2023-12) Phartiyal, Gopal Singh
The challenge of performing efficient and reliable crop classification with multisensor multitemporal (MSMT) images in mixed land cover scenarios i.e. presence of small land parcels (area < 20,000-meter square) of crops and other land covers such as built-up or grasslands, is significant. Specially in countries (ex. India) where diverse crops are practiced in small land parcels. This challenge can be addressed if deep neural network (DNN) based models can exploit all three i.e. spatial, spectral, and temporal information of a crop, present in the MSMT images, efficiently and effectively. Therefore, this study presents a novel DNN based model that exploits all three information in a synergistic fashion to achieve improved crop classification. At first, the model increases the significance of local spectral information via a strategy that creates versions of spectral band set wherein neighbourhood of spectral bands is permuted. Then, the model utilizes three-dimensional convolutions, in a time-distributed fashion, to extract local spectral-spatial features. Finally, the model utilizes bidirectional long short-term memory or LSTM-RNNs to extract the temporal information embedded in the time-distributed feature-space created after the convolutions. The developed model is trained and evaluated on Sentinel-1 and Sentinel-2 MSMT data to achieve a 6-class classification including two major crops grown in the region. One of the proposed models namely Perm-3D-CRNN-v1 showed a 97.77 % overall accuracy on test samples and reflected satisfactory on quantitative analysis. The localized spectral-spatial convolutions created prominent class-specific features whereas the bidirectional information flow in the recurrent layer improved the exploitation of crop-phenology type features making the model perform efficiently.
Synthetic imaging radar data generation in various clutter environments using novel uwb log-periodic antenna
(MDPI, 2024-12) Phartiyal, Gopal Singh
In short-range microwave imaging, the collection of data in real environments for the purpose of developing techniques for target detection is very cumbersome. Simultaneously, to develop effective and efficient AI/ML-based techniques for target detection, a sufficiently large dataset is required. Therefore, to complement labor-intensive and tedious experimental data collected in a real cluttered environment, synthetic data generation via cost-efficient electromagnetic wave propagation simulations is explored in this article. To obtain realistic synthetic data, a 3-D model of an antenna, instead of a point source, is used to include the coupling effects between the antenna and the environment. A novel printed scalable ultra-wide band (UWB) log-periodic antenna with a tapered feed line is designed and incorporated in simulation models. The proposed antenna has a highly directional radiation pattern with considerable high gain (more than 6 dBi) on the entire bandwidth. Synthetic data are generated for two different applications, namely through-the-wall imaging (TWI) and through-the-foliage imaging (TFI). After the generation of synthetic data, clutter removal techniques are also explored, and results are analyzed in different scenarios. Post-analysis shows evidence that the proposed UWB log-periodic antenna-based synthetic imagery is suitable for use as an alternative dataset for TWI and TFI application development, especially in training machine learning models.