Keshav D Singh

Dr. Keshav D Singh

Agriculture and Agri-Food Canada, Ottawa, Ontario, Canada
orcid

Research Focus & Metrics

Academic Career Length

15Years (2012-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
2 January 2021 in Canadian Journal of Remote Sensing
Verified
Journal Article
Closed access
Keshav D SinghKeshav D Singh
Keshav D Singh
HDHema S. N. Duddu
Hema S. N. Duddu
SVSally Vail
Sally Vail
IPIsobel Parkin
Isobel Parkin
SSSteve J. Shirtliffe
Steve J. Shirtliffe
Identification of optimal pod maturity stage in Canola is key for maximizing seed yield, quality and also an important phenotypic trait in crop improvement programs. The conventional method is via visual inspection of seed color change. Alternatively, hyperspectral sensors have potential to determine physiological status of the crops. Therefore, the objective of the study is to estimate canola seed maturity using field-based hyperspectral imaging. For this study, five canola genotypes (NAM-0, NAM-13, NAM-17, NAM-48, and NAM-76) were selected from an experiment of 56 populations. The experimental field was imaged using a UAV-mounted hyperspectral camera (400–1,000 nm) at five growth stages starting from pod formation to near-harvest maturity (BBCH-79(S1) to 88(S5)). For each genotype; pod and seed moisture were estimated on the same day of imaging. First-order derivative was conducted on reflectance data to determine optimal spectral wavebands. As a part of this study, a new vegetation index denoted “Canola-Pod-Maturity Index (CPMI)” was developed. CPMI was evaluated in comparison with four existing vegetation indices (mNDRE, PSRI, MCARI, WBI). CPMI showed a stronger relationship ( R2≃ 0.81–0.98 for pods and 0.66–0.85 for seeds) with pod and seed moisture for all the genotypes. Furthermore, the new index was able to find differences among genotypes with variable maturity times.

6 September 2024 in Sensors
Verified
Journal Article
Open access
PDF file
HWHongquan Wang
Hongquan Wang
Keshav D SinghKeshav D Singh
Keshav D Singh
HPHari Prasad Poudel
Hari Prasad Poudel
MNManoj Natarajan
Manoj Natarajan
PRPrabahar Ravichandran
Prabahar Ravichandran
BEBrandon Eisenreich
Brandon Eisenreich
Crop height and biomass are the two important phenotyping traits to screen forage population types at local and regional scales. This study aims to compare the performances of multispectral and RGB sensors onboard drones for quantitative retrievals of forage crop height and biomass at very high resolution. We acquired the unmanned aerial vehicle (UAV) multispectral images (MSIs) at 1.67 cm spatial resolution and visible data (RGB) at 0.31 cm resolution and measured the forage height and above-ground biomass over the alfalfa (Medicago sativa L.) breeding trials in the Canadian Prairies. (1) For height estimation, the digital surface model (DSM) and digital terrain model (DTM) were extracted from MSI and RGB data, respectively. As the resolution of the DTM is five times less than that of the DSM, we applied an aggregation algorithm to the DSM to constrain the same spatial resolution between DSM and DTM. The difference between DSM and DTM was computed as the canopy height model (CHM), which was at 8.35 cm and 1.55 cm for MSI and RGB data, respectively. (2) For biomass estimation, the normalized difference vegetation index (NDVI) from MSI data and excess green (ExG) index from RGB data were analyzed and regressed in terms of ground measurements, leading to empirical models. The results indicate better performance of MSI for above-ground biomass (AGB) retrievals at 1.67 cm resolution and better performance of RGB data for canopy height retrievals at 1.55 cm. Although the retrieved height was well correlated with the ground measurements, a significant underestimation was observed. Thus, we developed a bias correction function to match the retrieval with the ground measurements. This study provides insight into the optimal selection of sensor for specific targeted vegetation growth traits in a forage crop.

4 June 2025 in Sensors
Verified
Journal Article
Open access
PDF file
Shubham Subrot PanigrahiShubham Subrot Panigrahi
Shubham Subrot Panigrahi
Keshav D SinghKeshav D Singh
Keshav D Singh
PBParthiba Balasubramanian
Parthiba Balasubramanian
HWHongquan Wang
Hongquan Wang
MNManoj Natarajan
Manoj Natarajan
PRPrabahar Ravichandran
Prabahar Ravichandran
Dry bean, the fourth-largest pulse crop in Canada is increasingly impacted by climate variability, needing efficient methods to support cultivar development. This study investigates the potential of unmanned aerial vehicle (UAV)-based Light Detection and Ranging (LiDAR) and multispectral imaging (MSI) for high-throughput phenotyping of dry bean traits. Image data were collected across two dry bean field trials to assess plant height, lodging and seed yield. Multiple LiDAR-derived features accessing canopy height, crop lodging and digital biomass were evaluated against manual height measurements, visually rated lodging scale and seed yield, respectively. At the same time, three MSI-derived data were used to estimate seed yield. Classification- and regression-based machine learning models were used to estimate key agronomic traits using both LiDAR and MSI-based crop features. The canopy height derived from LiDAR showed a good correlation (R2 = 0.86) with measured plant height at the mid-pod filling (R6) stage. Lodging classification was most effective using Gradient Boosting, Random Forest and Logistic Regression, with R8 (physiological maturity stage) canopy height being the dominant predictor. For seed yield prediction, models integrating LiDAR and MSI outperformed individual datasets, with Gradient Boosting Regression Trees yielding the highest accuracy (R2 = 0.64, RMSE = 687.2 kg/ha and MAE = 521.6 kg/ha). Normalized Difference Vegetation Index (NDVI) at the R6 stage was identified as the most informative spectral feature. Overall, this study demonstrates the importance of integrating UAV-based LiDAR and MSI for accurate, non-destructive phenotyping in dry bean breeding programs.

31 January 2017 in International Journal of Remote Sensing
Verified
Journal Article
Closed access
Keshav D SinghKeshav D Singh
Keshav D Singh
DRDesikan Ramakrishnan
Desikan Ramakrishnan
From geological and planetary exploration perspectives, automated sub-pixel classification of hyperspectral data is the most difficult task as it involves blind unmixing with library spectra of minerals. In this study, we demonstrate a procedure involving spectral transformation and linear unmixing to achieve the above task. For this purpose, infrared spectra of rocks from the spectral library, field, and remotely sensed hyperspectral image cube were used. Potential spectra of minerals for unmixing rock spectra were drawn from the library based on similarity of absorption features measured using Pearson correlation coefficient. Eight transformation techniques namely, first derivative, fast Fourier transform, discrete wavelet transform, Hilbert–Huang transform, crude low pass filter, S-transform, binary encoding, spectral effective peak matching, and two sparsity-based techniques (orthogonal matching pursuit, sparse unmixing via variable splitting, and augmented Lagrangian) were evaluated. Subsequently, minerals identified by above techniques were unmixed by linear mixture model (LMM) to decipher mineralogical composition and abundance. Results of LMM achieved using fully constrained least-square-estimation-based quadratic programming optimization approach were evaluated by conventional procedures such as X-ray diffraction and microscopy. In the case of image cube, endmembers derived using minimum noise fraction and pixel purity index were subjected to above procedure. It is evident that the discrete-wavelet-transformation-based approach produced excellent and meaningful results due to its flexibility in scaling the data and capability to handle noisy spectra. It is interesting to note that the adopted procedure could perform sub-pixel classification of image cube automatically and identify predominance of dolomite in limestone and sodium in alunite based on subtle differences in absorption positions.

28 June 2024 in IEEE Transactions on AgriFood Electronics
Verified
Journal Article
Closed access
SPStephany Solange Valarezo Plaza
Stephany Solange Valarezo Plaza
JTJulio Torres-Tello
Julio Torres-Tello
Keshav D SinghKeshav D Singh
Keshav D Singh
SSSteve J. Shirtliffe
Steve J. Shirtliffe
SDS. Deivalakshmi
S. Deivalakshmi
SKSeokbum Ko
Seokbum Ko
The escalating global demand for food, coupled with challenges in sustaining crop production, deteriorating ocean health, and depleting natural resources, underscores the critical role of agricultural technology. This article addresses the imperative of developing an optimal deep-learning model for predicting canola crop yield using hyperspectral images captured by drone flights. Our primary objective is to identify the most efficient model in terms of performance and size, considering the storage limitations on edge devices like Raspberry Pi 4 (RPi4). We start with the baseline 1D _ CNN model, which achieves an $R^{2}$ score of 0.82, and compress it into the proposed fs_model (fp32). To achieve the compression, we apply pruning through sparsity and feature selection using SHAP values. Further reduction in model size is accomplished by quantizing the weights of the proposed model to a lower precision, such as int16. This combined approach substantially decreases the proposed model's size by approximately 92.6% and inference time by approximately ×9013 in comparison to the baseline 1D _ CNN model. In addition, we propose the novel fsp_model posit(8,3) that uses posit quantization to further reduce the computation requirements compared to the proposed fs_model (int16). Our findings indicate that the utilization of posit numbers enables us to shrink the model size to 94% of the original base model, while only reducing the $R^{2}$ score by 5.7%.

1 July 2012
Verified
Conference Paper
Closed access
Keshav D SinghKeshav D Singh
Keshav D Singh
DRDesikan Ramakrishnan
Desikan Ramakrishnan
LMLalu Mansinha
Lalu Mansinha
One of the tedious and time consuming tasks related to hyperspectral data analysis is the identification of library candidates for spectral unmixing. In this study, we evaluated the relevance of different transformation procedures such as First Derivative (FD), Fast Fourier Transform (FFT), Discrete Wavelet Transform (DWT), Hilbert-Huang Transform (HHT) and S-transform (ST) in automated retrieval of library endmembers for linear spectral unmixing. The spectral similarity between the target and library candidates were estimated using Pearson's Correlation Coefficient (PCC) and student t-test based approach. Subsequently, these endmembers are used to estimate the fractional abundances by Fully Constrained Least Square Estimation (FCLSE) based Quadratic Programming (QP) optimization approach. The match between the target and modeled spectrum was calculated based on Root Mean Squared Error (RMSE) and spectral similarity scores estimated using Spectral Angle Mapper (SAM). In addition to RMSE and SAM scores, the simulation processing time and appropriateness of identified endmembers are considered to estimate the effectiveness of each transformation procedure. It is observed that DWT, HHT and ST based approaches are more efficient in identifying correct library endmembers than the FD and FFT based approaches.