International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 12 Issue: 12 | Dec 2025
p-ISSN: 2395-0072
www.irjet.net
De noise of Plant Images Using Generative Artificial Intelligence and Transfer Learning Md. Gous Azam1,Dr. Ashok Kumar2 1Research Scholar, Department of Computer Science, V.K.S.U., Ara 2Professor, Ex. Head Department of Physics, V.K.S.U., Ara
---------------------------------------------------------------------***----------------------------------------------------------------------
Abstract— In recent years, owing to the rapid evolution
allows better management and timely interventions to prevent substantial
of machine learning, especially deep learning, and its outstanding performance in the domain of image processing an increasing number of scholars are turning to convolutional neural networks (CNNs) to address image de noise challenges. To tackle image de noise issue, this paper introduces the implementation of a deep Generative Adversarial Network for image de-hazing. Unlike traditional de-hazing methods that rely on per-pixel loss, our approach leverages a perceptual loss function. This function captures high-level image features by utilizing pre-trained models from Image Net, effectively mitigating the shortcomings associated with per-pixel loss functions, particularly their sensitivity to minor variations in single pixels, even when the images are perceptually similar.
crop losses. Therefore, early-stage disease detection remains essential to protect plants from harmful diseases and ensure healthier crops. Farmers require continuous monitoring of their plants to protect them from harmful diseases. However, this process is often expensive and time-consuming in real-world agriculture. To ensure accurate disease detection, image pre-processing plays an important role in improving the quality of images, which helps achieve better diagnostic outcomes. For effective disease identification through automated systems, highquality image processing is essential. Input images often suffer from excess noise due to environmental disturbances, and if the data is not properly cleaned, reliable results cannot be obtained [5][6]. When distortionfree images are provided as input, diseases can be more effectively detected, allowing for timely and appropriate interventions. However, many existing research studies place less emphasis on the pre-processing stage, leading to suboptimal results. Ineffective pre-processing leads to various problems including degraded image quality, suppressed performance due to excessive noise, loss of crucial information, increased processing time and inaccurate analysis. To ensure effective automated analysis of plant diseases, it is essential that the input images are of high quality. High-quality images allow for a more detailed examination of the data, allowing the extraction of critical information. This, in turn, helps identify diseases early and implement appropriate solutions in a timely manner [7][8].Image de noising is a crucial aspect of image processing and computer vision dealing with noise during image-taking, transmission, or processing. Noise is a factor that results in a decrease in image quality which directly disturbs the accuracy of image-based analysis and it further affects the recognition, segmentation, and tracking tasks. Therefore, image de noising has been a major research topic, which has produced a number of techniques to deal with these challenges [9][10]. The traditional methods of de noising have some shortcomings like demanding high computational power, deficiency of the adaptation to various noise types and levels, and challenges in preserving image details and textures. In recent years, deep learning-based approaches have emerged as a promising solution, achieving state-of-the-art performance in view of both quantitative metrics and visual quality. The effectiveness of deep learning methods
Keywords— Image De noise, Plant Village, Apple, Generative Artificial Intelligence, Transfer Learning
1. INTRODUCTION Agriculture plays an important role in India's economy, with nearly half of the population depending on farming for their livelihood. The main aim of agricultural research is to improve the productivity and quality of food while minimizing costs and maximizing profits. The success of agricultural production relies on the effective integration of factors such as seeds, soil, and agrochemicals. Achieving high-quality and valuable products requires stringent quality control measures [1][2]. To ensure efficient production and surplus yields, continuous enhancement of agricultural practices and regular product value assessments remain essential. Plant diseases, however, pose significant challenges by disrupting key processes like photosynthesis, pollination, germination, transpiration, and fertilization, which hinders the plant's overall health and productivity. Unforeseen environmental conditions often lead to the emergence of diseases caused by pathogens such as bacteria, viruses, and fungi. In India, a wide variety of crops are grown across different regions each with its own unique climate and seasonal variations. Early detection of diseases in plant leaves is particularly crucial in agriculture. Plant leaf diseases are a major factor that contributes to reduced crop yields worldwide. Farmers frequently face significant problems in controlling crop diseases, especially at advanced stages [3][4]. Accurate disease diagnosis is vital in agriculture as it
© 2025, IRJET
|
Impact Factor value: 8.315
|
ISO 9001:2008 Certified Journal
|
Page 1042