Comparative Evaluation of Modified Wasserstein GAN-GP and State-of-the-Art GAN Models for Synthesizing Agricultural Weed Images in RGB and Infrared Domain

This study investigates the application of modified Wasserstein Generative Adversarial Networks with Gradient Penalty (WGAN-GP) to generate synthetic RGB and infrared (IR) datasets to meet the annotation requirements for wild radish (Raphanus raphanistrum). The RafanoSet dataset was used for evaluat...

Full description

Saved in:
Bibliographic Details
Main Authors: Shubham Rana, Matteo Gatti
Format: Article
Language:English
Published: Elsevier 2025-06-01
Series:MethodsX
Subjects:
Online Access:http://www.sciencedirect.com/science/article/pii/S2215016125001554
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849471495996178432
author Shubham Rana
Matteo Gatti
author_facet Shubham Rana
Matteo Gatti
author_sort Shubham Rana
collection DOAJ
description This study investigates the application of modified Wasserstein Generative Adversarial Networks with Gradient Penalty (WGAN-GP) to generate synthetic RGB and infrared (IR) datasets to meet the annotation requirements for wild radish (Raphanus raphanistrum). The RafanoSet dataset was used for evaluation. Traditional WGAN models struggle with vanishing gradients and poor convergence, affecting data quality. Customizations in WGAN-GP improved synthetic image quality, especially in maintaining SSIM for RGB datasets. However, generating high-quality IR images remains challenging due to spectral complexities, with lower SSIM scores. Architectural enhancements including transposed convolutions, dropout, and selective batch normalization improved SSIM scores from 0.5364 to 0.6615 for RGB and from 0.3306 to 0.4154 for IR images. This study highlights the customized model's key features: • Produces a 128 × 7 × 7 tensor, optimizes feature map size for subsequent layers, with two layers using 4 × 4 kernels and 128 and 64 filters for upsampling. • Uses 3 × 3 kernels in all convolutional layers to capture fine-grained spatial features, incorporates batch normalization for training stability, and applies dropout to reduce overfitting and improve generalization.
format Article
id doaj-art-9efe970e522045fd8fb401a23e2d2b43
institution Kabale University
issn 2215-0161
language English
publishDate 2025-06-01
publisher Elsevier
record_format Article
series MethodsX
spelling doaj-art-9efe970e522045fd8fb401a23e2d2b432025-08-20T03:24:48ZengElsevierMethodsX2215-01612025-06-011410330910.1016/j.mex.2025.103309Comparative Evaluation of Modified Wasserstein GAN-GP and State-of-the-Art GAN Models for Synthesizing Agricultural Weed Images in RGB and Infrared DomainShubham Rana0Matteo Gatti1Corresponding author.; Department of Sustainable Crop Production, Università Cattolica del Sacro Cuore, Via Emilia Parmense 84, Piacenza, 29122, ItalyDepartment of Sustainable Crop Production, Università Cattolica del Sacro Cuore, Via Emilia Parmense 84, Piacenza, 29122, ItalyThis study investigates the application of modified Wasserstein Generative Adversarial Networks with Gradient Penalty (WGAN-GP) to generate synthetic RGB and infrared (IR) datasets to meet the annotation requirements for wild radish (Raphanus raphanistrum). The RafanoSet dataset was used for evaluation. Traditional WGAN models struggle with vanishing gradients and poor convergence, affecting data quality. Customizations in WGAN-GP improved synthetic image quality, especially in maintaining SSIM for RGB datasets. However, generating high-quality IR images remains challenging due to spectral complexities, with lower SSIM scores. Architectural enhancements including transposed convolutions, dropout, and selective batch normalization improved SSIM scores from 0.5364 to 0.6615 for RGB and from 0.3306 to 0.4154 for IR images. This study highlights the customized model's key features: • Produces a 128 × 7 × 7 tensor, optimizes feature map size for subsequent layers, with two layers using 4 × 4 kernels and 128 and 64 filters for upsampling. • Uses 3 × 3 kernels in all convolutional layers to capture fine-grained spatial features, incorporates batch normalization for training stability, and applies dropout to reduce overfitting and improve generalization.http://www.sciencedirect.com/science/article/pii/S2215016125001554Modified Wasserstein Generative Adversarial Network with Gradient Penalty
spellingShingle Shubham Rana
Matteo Gatti
Comparative Evaluation of Modified Wasserstein GAN-GP and State-of-the-Art GAN Models for Synthesizing Agricultural Weed Images in RGB and Infrared Domain
MethodsX
Modified Wasserstein Generative Adversarial Network with Gradient Penalty
title Comparative Evaluation of Modified Wasserstein GAN-GP and State-of-the-Art GAN Models for Synthesizing Agricultural Weed Images in RGB and Infrared Domain
title_full Comparative Evaluation of Modified Wasserstein GAN-GP and State-of-the-Art GAN Models for Synthesizing Agricultural Weed Images in RGB and Infrared Domain
title_fullStr Comparative Evaluation of Modified Wasserstein GAN-GP and State-of-the-Art GAN Models for Synthesizing Agricultural Weed Images in RGB and Infrared Domain
title_full_unstemmed Comparative Evaluation of Modified Wasserstein GAN-GP and State-of-the-Art GAN Models for Synthesizing Agricultural Weed Images in RGB and Infrared Domain
title_short Comparative Evaluation of Modified Wasserstein GAN-GP and State-of-the-Art GAN Models for Synthesizing Agricultural Weed Images in RGB and Infrared Domain
title_sort comparative evaluation of modified wasserstein gan gp and state of the art gan models for synthesizing agricultural weed images in rgb and infrared domain
topic Modified Wasserstein Generative Adversarial Network with Gradient Penalty
url http://www.sciencedirect.com/science/article/pii/S2215016125001554
work_keys_str_mv AT shubhamrana comparativeevaluationofmodifiedwassersteingangpandstateoftheartganmodelsforsynthesizingagriculturalweedimagesinrgbandinfrareddomain
AT matteogatti comparativeevaluationofmodifiedwassersteingangpandstateoftheartganmodelsforsynthesizingagriculturalweedimagesinrgbandinfrareddomain