A Deep Learning Approach for Diagnosing Nutrient Deficiencies in Okra Leaves

Fuente: Zenodo
Saved in:
Bibliographic Details
Main Authors: Dipankar Das, Uzzal Sharma, Gypsy Nandi
Format: Recurso digital
Published: Zenodo 2025
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866902285674610688
author Dipankar Das
Uzzal Sharma
Gypsy Nandi
author_facet Dipankar Das
Uzzal Sharma
Gypsy Nandi
contents <p>Precision farming and production optimization depend on accurate nutrient deficit <br>(ND) diagnosis.  In order to diagnose various problems in okra leaves, this study <br>proposes a repeatable and comprehensible deep-learning pipeline that combines <br>segmentation, adaptive picture improvement, handcrafted deep feature fusion, and <br>a lightweight convolutional neural network (CNN) classifier. <br>By simultaneously adjusting Gaussian and Poisson noise weights, a Dual Adaptive <br>Weighted Kalman Filter (DAWKF) reduces noise in input images.  Segmentation is <br>carried out via an Adaptive Weighted Optimizer with Otsu Thresholding (AWO-AOT) <br>using bio-inspired adaptive migration methods that improve local thresholds.  Local <br>Binary Patterns (LBP), FAST, and SIFT descriptors are used in Adaptive Fused <br>Features (AFF), which feed a SIFT-guided CNN to learn spatial-texture <br>representations. <br>2,030 agronomist-verified okra-leaf photos from various field sources make up the <br>extended dataset. They are divided into 70/15/15 sections for training, validation, <br>and testing, and they are assessed using five-fold cross-validation.  The suggested <br>framework outperformed BM3D and PSO-Otsu baselines with 98.6 ± 0.3% accuracy <br>and statistically significant gains (p < 0.05) in PSNR (+1.8 dB) and SSIM (+0.014). <br>Grad-CAM explainability studies revealed important symptom regions associated <br>with necrosis and leaf chlorosis.  The technique provides a repeatable mechanism <br>for real-field agronomic decision assistance and is generalizable across lighting, <br>camera, and field conditions.</p>
format Recurso digital
id zenodo_https___doi_org_10_5281_zenodo_18066359
institution Zenodo
language
publishDate 2025
publisher Zenodo
record_format zenodo
spellingShingle A Deep Learning Approach for Diagnosing Nutrient Deficiencies in Okra Leaves
Dipankar Das
Uzzal Sharma
Gypsy Nandi
<p>Precision farming and production optimization depend on accurate nutrient deficit <br>(ND) diagnosis.  In order to diagnose various problems in okra leaves, this study <br>proposes a repeatable and comprehensible deep-learning pipeline that combines <br>segmentation, adaptive picture improvement, handcrafted deep feature fusion, and <br>a lightweight convolutional neural network (CNN) classifier. <br>By simultaneously adjusting Gaussian and Poisson noise weights, a Dual Adaptive <br>Weighted Kalman Filter (DAWKF) reduces noise in input images.  Segmentation is <br>carried out via an Adaptive Weighted Optimizer with Otsu Thresholding (AWO-AOT) <br>using bio-inspired adaptive migration methods that improve local thresholds.  Local <br>Binary Patterns (LBP), FAST, and SIFT descriptors are used in Adaptive Fused <br>Features (AFF), which feed a SIFT-guided CNN to learn spatial-texture <br>representations. <br>2,030 agronomist-verified okra-leaf photos from various field sources make up the <br>extended dataset. They are divided into 70/15/15 sections for training, validation, <br>and testing, and they are assessed using five-fold cross-validation.  The suggested <br>framework outperformed BM3D and PSO-Otsu baselines with 98.6 ± 0.3% accuracy <br>and statistically significant gains (p < 0.05) in PSNR (+1.8 dB) and SSIM (+0.014). <br>Grad-CAM explainability studies revealed important symptom regions associated <br>with necrosis and leaf chlorosis.  The technique provides a repeatable mechanism <br>for real-field agronomic decision assistance and is generalizable across lighting, <br>camera, and field conditions.</p>
title A Deep Learning Approach for Diagnosing Nutrient Deficiencies in Okra Leaves
url https://doi.org/10.5281/zenodo.18066359