article · Engineering Technology & Applied Science Research
Deep learning techniques can improve the detection of specific tomato conditions, including healthy fruit, blossom end rot, splitting rotation, and sun-scaled rotation. By evaluating two lightweight object detection architectures, YOLOv5l and YOLOv8l, against a custom dataset of tomato diseases, baselines were established prior to expanding the data. Data augmentation methods sourced from Roboflow were subsequently introduced to expose the systems to diverse lighting, poses, and background settings. Re-training both models on this augmented dataset led to a substantial increase in diagnostic accuracy across all four categories. Overall, YOLOv8l consistently demonstrated marginally superior accuracy relative to YOLOv5l, an advantage that was especially pronounced when background imagery was omitted from the assessment.
Tomato crops are susceptible to various conditions that damage yield and fruit quality. Using lightweight computer vision models to accurately identify specific physical defects, such as rot or splitting, provides a practical basis for automated crop monitoring. Demonstrating that data augmentation improves detection under varied lighting and backgrounds helps make automated tools more reliable in practical agricultural settings.
This research provides applied and tested models that could enable automated crop monitoring tools or sorting systems for tomato growers and agricultural technology developers. Because the models tested are lightweight, they could potentially be integrated into mobile or edge-computing devices for field scouting, although further deployment testing outside the custom dataset would be required before market adoption.
AI-generated from the published abstract. Always read the original work before citing.
This study delves into the application of deep learning for precise tomato disease detection, focusing on four crucial categories: healthy, blossom end rot, splitting rotation, and sun-scaled rotation. The performance of two lightweight object detection models, namely YOLOv5l and YOLOv8l, was compared on a custom tomato disease dataset. Initially, both models were trained without data augmentation to establish a baseline. Subsequently, diverse data augmentation techniques were obtained from Roboflow to significantly expand and enrich the dataset content. These techniques aimed to enhance the models' robustness to variations in lighting, pose, and background conditions. Following data augmentation, the YOLOv5l and YOLOv8l models were re-trained and their performance across all disease categories was meticulously analyzed. After data augmentation, a significant improvement in accuracy was observed for both models, highlighting its effectiveness in bolstering the models' ability to accurately detect tomato diseases. YOLOv8l consistently achieved slightly higher accuracy compared to YOLOv5l, particularly when excluding background images from the evaluation.
This page summarises published work. The authoritative version sits with the publisher.
DOI: 10.48084/etasr.7262
Is something wrong with this record? Report it or request removal.
Discussion
Have you built on this work, tried to replicate it, or seen it applied in practice? Share what you know. Verified researchers and MARATTO™ domain experts can open a discussion, and any member can reply. Contributions are reviewed before they appear.
No discussion yet. Open the first thread.
New to MARATTO™? Create a free account.
