Aug 2026· International Conference Electronic Systems, Signal Processing and Computing Technologies [ICESC-]· pp. 1951-1956· 0 citations· 27 references
TL;DR
An assessment of a DeiT-III Small transformer backbone for 5-class groundnut leaf disease classification with minimal pre-processing and a controlled experimental protocol gives a good leakage-safe groundnut disease classification under minimal preprocessing.
Abstract
Modern vision transformer backbones have shown strong representation learning ability, but their direct effectiveness on small agricultural disease datasets requires careful leakage-safe benchmarking. This paper presents an assessment of a DeiT-III Small transformer backbone for 5-class groundnut leaf disease classification with minimal pre-processing and a controlled experimental protocol. No extra attention modules or domain-specific preprocessing are introduced in the study, and the capability of a modern transformer backbone under resizing and normalization is examined. The dataset contains 1,720 groundnut leaf images across five classes: Alternaria Leaf Spot, Healthy, Leaf Spot (Early and Late), Rosette, and Rust. To mitigate optimistic bias, exact and perceptual duplicates were clustered, augmentation was applied to the training partition only, the validation macro-F1 was used to select models, and the test set was kept for final testing. DeiT-III Small scored 98.13% accuracy, 98.22% macro-F1, 98.12% weighted-F1, and 98.59% balanced accuracy on the held-out test partition of 267 images. The model also obtained Cohen’s kappa of 97.44% and Matthews correlation coefficient of 97.46%. Five cases were misclassified, primarily the visually similar lesion classes. Disease-sensitive regions were inspected using the saliency maps of the transformer. The results obtained give a good leakage-safe groundnut disease classification using the transformer under minimal preprocessing.
Accurate and early detection of citrus leaf diseases is essential for maintaining lemon crop health and yield. This work presents a comparative evaluation of four deep learning architectures — CNN, CNN+LSTM, ResNet50, and DenseNet121 — for multi-class classification of lemon leaf diseases using image-based data. A stan...
Kunal Ramesh Ahire· Journal of Intelligent Decis...· 0 citations
Conventional image augmentation is widely used to improve model generalization; however, its benefit in fine-grained plant disease classification can be overstated when duplicated or near-duplicated images cross the train–test boundary. This study tests whether conventional augmentation remains beneficial for sugarcane...
J. Budiasto, H. Jayawardana, Tri Kustanti Rahayu et al.· Journal of Information Syste...· 0 citations
Tomato leaf diseases remain a major cause of yield loss in one of the most widely cultivated vegetable crops, and manual, expert-dependent diagnosis is slow, inconsistent, and often inaccessible in resource-limited farming regions. This study proposes a fine-tuned Swin Transformer V2 model for automated tomato leaf dis...
Gopal Bhojak, Ruchi Dave, Pratibha Soni· International Journal of Inn...· 0 citations
It is suggested that lightweight feature recalibration can enhance a compact ViT without substantial parameter overhead, however, the limited dataset size and absence of an independent external test set require cautious interpretation and further validation across diverse field conditions.
Pravangkar Boruah, R. Bania· Archives of Current Research...· 0 citations
This study presents a reliability-oriented framework for evaluating whether high classification performance in rice leaf disease recognition is accompanied by spatially faithful and generalizable explanations. The primary dataset comprised 4,684 images from three disease classes: bacterial blight, brown spot, and leaf...
Sujeet Kumar, Prajeet Sharma· International Journal of Ele...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.