Plant diseases pose a significant threat to agricultural productivity and global food security, accounting for 70-80% of crop losses worldwide. Traditional detection methods rely heavily on expert visual inspection, which is time-consuming, labour-intensive, and often impractical for large-scale farming operations. In this paper, we present PlantDiseaseNet-RT50, a novel fine-tuned deep learning architecture based on ResNet50 for automated plant disease detection. Our model features strategically unfrozen layers, a custom classification head with regularization mechanisms, and dynamic learning rate scheduling through cosine decay. Using a comprehensive dataset of distinct plant disease categories across multiple crop species, PlantDiseaseNet-RT50 achieves exceptional performance with approximately 98% accuracy, precision, and recall. Our architectural modifications and optimization protocol demonstrate how targeted fine-tuning can transform a standard pretrained model into a specialized agricultural diagnostic tool. We provide a detailed account of our methodology, including the systematic unfreezing of terminal layers, implementation of batch normalization and dropout regularization and application of advanced training techniques. PlantDiseaseNet-RT50 represents a significant advancement in AI-driven agricultural tools, offering a computationally efficient solution for rapid and accurate plant disease diagnosis that can be readily implemented in practical farming contexts to support timely interventions and reduce crop losses.
Agriculture, a primary source of food supply, is expanding constantly. Agriculture is not only a vital core occupation but also significantly contributes to the expansion of the national economy.
Plant diseases incur significant costs for farmers. Numerous diseases can adversely affect various components of a plant’s anatomy, including its leaves, stems, seeds, and fruit. The body of a plant gets segmented into various portions due to specific illnesses. Leaves can be regarded as the fundamental component of a plant. Consequently, diseases lead to complete or partial crop failures, diminishing food supply and exacerbating food insecurity [1].
Food crops are significantly harmed by fungi, bacteria, viruses, high winds, adverse climatic conditions, and drought. Plant diseases constitute 70-80% of plant losses. A multitude of fungal and bacterial diseases impact most plants. Plant diseases are also induced by climatic conditions and the exponential growth of the population. Meticulous examination of the leaves will be necessary to identify the affliction [2].
In this paper, we introduce PlantDiseaseNet-RT50, a novel fine-tuned deep learning architecture designed specifically for high-accuracy plant disease detection. By systematically optimizing a ResNet50 backbone through strategic layer unfreezing, custom classification head design, and advanced training methodologies, we demonstrate how architectural refinement can dramatically enhance performance in specialized agricultural applications. The development of PlantDiseaseNet-RT50 addresses critical challenges in automated plant disease detection, including feature extraction efficiency, computational resource requirements, and the ability to generalize across diverse plant pathologies.
Plant diseases are typically categorised into three types: fungal diseases, bacterial diseases, and viral illnesses. Examples of plant diseases include leaf rust, downy mildew, brown spot, powdery mildew, and bacterial blight. Fungal diseases are prevalent in plants, accounting for almost 80 percent of all plant illnesses. Bacteria and viruses can also induce severe diseases in plants; however, the incidence of such diseases is significantly lower compared to those caused by fungi. Certain insects are also implicated in many plant diseases. Plant diseases can be categorised as biotic and abiotic. Biotic disorders exhibit certain visual signs. These visual manifestations are regarded as signs. Bacterial exudate and fungal proliferation are examples of signs observed on plant leaves. Abiotic diseases are conditions that do not often propagate. No signs will be evident in abiotic disorders. Abiotic diseases result from nutritional deficiencies, air pollution, water pollution, and similar factors. Abiotic diseases are challenging to identify due to the absence of observable signs. In contrast to abiotic diseases, biotic diseases have the ability to propagate, necessitating the implementation of stringent measures to combat them [3].
Convolutional neural networks (CNNs) are a deep learning methodology. They excel in image recognition and processing applications. CNN are especially adept at plant disease identification due to their proficiency in image processing and feature extraction. CNNs may be trained on extensive datasets of plant photos, both diseased and healthy, enabling them to discern nuanced patterns and characteristics of various illnesses [7].
Convolutional neural networks are a type of deep neural networks engineered solely for image recognition and processing applications. Inspired by the visual processing of the human brain, they proficiently capture hierarchical elements and patterns in images. A fundamental CNN architecture comprises an input layer, several convolutional layers followed by pooling layers, and fully connected (dense) layers culminating in the output layer. The convolutional layers utilise filters on the input to extract features, whereas the pooling layers reduce the spatial dimensions, resulting in more efficient processing. The thick layers subsequently execute classification utilising the extracted features [11].
Convolutional layers, which extract features such as edges, textures, and forms from the input image through the application of filters, are the essential elements of a CNN. The feature maps are down sampled to diminish spatial dimensions while preserving the most significant data, subsequent to the pooling layers’ processing of the convolutional layers’ output [8]. The output from the pooling layers is further processed by the dense layers to categorise or predict the image.
Mohanty et al. demonstrated that a finetuned deep CNN can achieve > 90% accuracy, validating the feasibility of smartphone-assisted disease diagnosis when data are controlled and curated. Ferentinos broadened the scope to 58 plant-disease classes, finding top models exceeding 95% accuracy and reinforcing that modern CNNs can scale across many crops and patho
This content is AI-processed based on open access ArXiv data.
