Enhancing Cross-Grade Generalization in Banana Variety Classification via Hybrid CNN-Transformers and Knowledge Self-Distillation

Khush Kataruka; Mohammad Muzammila; Zahid Akhtar; Raghavendra Ramachandra; Praveen Kumar Chandaliya

cs.CVOct 21, 2026

Enhancing Cross-Grade Generalization in Banana Variety Classification via Hybrid CNN-Transformers and Knowledge Self-Distillation

Authors: Khush Kataruka, Mohammad Muzammila, Zahid Akhtar, Raghavendra Ramachandra, Praveen Kumar Chandaliya

Organizations: Department of Artificial Intelligence, Sardar Vallabhbhai National Institute of Technology, Surat, Gujarat, India. · Electrical and Computer Engineering Department, State University of New York Polytechnic Institute, Utica, New York, USA. · Department of Information Security and Communication Technology, Norwegian University of Science and Technology, Gjovik, Norway.

Abstract

In recent years, automated fruit classification systems have become a more critical component in precision agriculture for efficient yield estimation, quality control, and supply chain optimization. However, they still face significant challenges under varying quality, fruit maturity, illumination, texture degradation, morphological attributes, domain shift, and post-harvest handling conditions. In this study, we address the challenging issue of cross-grade banana variety classification. This is a real-world setting where models that have been trained on specific quality grades must generalize to unseen grades to handle significant domain shift. Under a rigorous Leave-One-Grade-Out evaluation protocol, we present a thorough comparative analysis of CNNs, ViTs, and dual-stream hybrid architectures using a large-scale dataset of 31,678 banana images with three varieties and three quality grades. Specifically, to address the drawbacks of single-paradigm frameworks, we present a dual-stream Swin ResNet hybrid framework enhanced with KL-divergence-based self-distillation to align probabilistic feature representations across the CNN and Transformer streams. Comprehensive experimental analyzes show that the presented hybrid framework with KL-divergence based self-distillation achieves better overall performance, i.e., it attains a mean accuracy of 81.25% and significantly surpasses CNN and Transformer baseline models. Qualitative analysis further confirms that the hybrid framework in this study improves discriminative feature transferability, improves attention localization, and stabilizes optimization.

Figures & tables

Variety	Count	Percentage
Pachbale	11,512	36.34%
Rasbale	10,112	31.92%
Elakki-bale	10,054	31.74%
Total	31,678	100.00%

Table 1: Overall Dataset Statistics.

Grade	Variety	Count	Percentage	Grade Total
Grade 1 (Heavily blemished)	Pachbale	3,807	36.56%	10,412
	Rasbale	3,589	34.47%
	Elakki-bale	3,016	28.97%
Grade 2 (Moderate defects)	Pachbale	3,944	36.42%	10,830
	Elakki-bale	3,571	32.97%
	Rasbale	3,315	30.61%

Table 2: Variety Distribution Across Quality Grades

Figure 1: Illustration of the KL Divergence-based distillation mechanism used to align feature distributions.

Experiment	Training Grades	Test Grade	Train Size	Test Size
1	Grade 2 + Grade 3	Grade 1	19,139	10,412
2	Grade 1 + Grade 3	Grade 2	18,763	10,830
3	Grade 1 + Grade 2	Grade 3	19,117	10,436

Table 3: Experimental Configuration for Leave-One-Grade-Out (LOGO) Cross-Validation

Baselines
Model	Mean Acc.	Std Acc.	Mean F1	Best	Worst
ResNet50	0.6945	$\pm$ 0.0710	0.6597	0.7760	0.6465
EfficientNet-B1	0.7151	$\pm$ 0.0390	0.6837	0.7602	0.6922
ConvNeXt-Tiny	0.7413	$\pm$ 0.1155	0.6999	0.8711	0.6498
DenseNet-121	0.7858	$\pm$ 0.0536	0.7752	0.8362	0.7295
ViT-B/16	0.7581	$\pm$ 0.0940	0.7457	0.8338	0.6528

Table 4: Comprehensive Performance Evaluation of CNN-Based Backbones and Hybrid Architectures under KL Divergence Settings (No Augmentation)

Model	Test Grade	Accuracy	Precision	Recall	F1-Score
ResNet50	Grade1	0.6465	0.6511	0.6465	0.5907
	Grade2	0.6609	0.6581	0.6609	0.6143
	Grade3	0.7760	0.7893	0.7760	0.7741
EfficientNet-B1	Grade1	0.6930	0.7332	0.6930	0.6432
	Grade2	0.6922	0.7112	0.6922	0.6539
	Grade3	0.7602	0.7682	0.7602	0.7540

Table 5: Performance Comparison Across Test Grades

Model	Train Grade	Test Grade	Accuracy	Precision	Recall	F1-Score
Swin-ResNet (KL)	Grade1	Grade2	0.6044	0.4968	0.6044	0.5120
	Grade1	Grade3	0.7756	0.7972	0.7756	0.7643
	Grade2	Grade1	0.6177	0.4236	0.6177	0.4981
	Grade2	Grade3	0.7220	0.8047	0.7220	0.6834
	Grade3	Grade1	0.7108	0.7560	0.7108	0.6884
	Grade3	Grade2	0.9148	0.9307	0.9148	0.9163

Table 6: Cross-Grade Generalization Results under Train-on-One-Grade (T1G) Protocol

Reference Baselines (Independent Training)
Ablation Study	G1 Acc.	G2 Acc.	G3 Acc.	Mean Acc. $\pm$ Std	Mean F1	Best Acc.	Worst Acc.
Swin-Tiny (Teacher only)	0.6654	0.7980	0.7249	0.7294 $\pm$ 0.0664	0.7118	0.7980	0.6654
ResNet50 (Student only)	0.6465	0.6609	0.7760	0.6945 $\pm$ 0.0710	0.6597	0.7760	0.6465
Ablation Variants (Isolating Components)
w/o Adaptor (Dual CE)	0.6370	0.8111	0.8719	0.7733 $\pm$ 0.1219	0.7577	0.8719	0.6370
Adaptor only (Single CE)	0.6274	0.7507	0.8621	0.7467 $\pm$ 0.1174	0.7222	0.8621	0.6274

Table 7: Ablation Study: Grade-wise and Component-wise Performance Analysis

Panel of Figure 2 — Figure 2: Confusion Matrices for Swin-ResNet (KL) across all test grades. The strong diagonal even in Grade 1 demonstrates superior feature robustness compared to baseline CNNs.

Panel of Figure 3 — Figure 3: Training Dynamics Matrix. The left column shows the cross-entropy loss profiles, where the close alignment of training (blue) and validation (red) curves indicates minimal overfitting. The right column illustrates the rate of loss change, where the rapid settling near zero ( $\Delta L\approx 0$ ) confirms stable convergence without significant oscillations, even in the challenging Grade 1 scenario.

Model	Grade1 Test	Grade2 Test	Grade3 Test
ResNet50	25	19	12
EfficientNet-B1	19	18	17
Swin-Tiny	21	24	16
Swin-ResNet(Standard)	10	24	9
Swin-ResNet(KL)	8	8	10
Swin-EfficientNet(KL)	9	12	16

Table 8: Training Epochs Until Early Stopping

Panel of Figure 4 — Figure 4: Grade 1 Success vs. Failure Case Study.

Panel of Figure 5 — Figure 5: Qualitative comparison of correct and incorrect predictions for Grade 2 samples, highlighting differences in model attention and feature localization.

Panel of Figure 6 — Figure 6: Qualitative comparison of correct and incorrect predictions for Grade 3 samples, illustrating the model?s localization capability and challenges in fine-grained discrimination.

Panel of Figure 7 — Figure 7: Student vs. Teacher Attention Matrix. A 2 $\times$ 3 comparison across all quality grades. In each stacked image, the middle strip represents the Student’s (ResNet50) attention. The strong overlap between the Student’s focus and the relevant fruit regions indicates effective feature learning from the Teacher, even though the target layer is strictly from the ResNet50 backbone.

Grade	True Class	Sample ID	Grad-CAM IoU	Score-CAM IoU
Grade 1	Rasbale	9480	0.268	0.114
Grade 1	Elakki-bale	1623	0.313	0.202
Grade 2	Rasbale	10169	0.362	0.221
Grade 2	Elakki-bale	1905	0.375	0.017
Grade 3	Elakki-bale	2972	0.307	0.217
Grade 3	Rasbale	9392	0.176	0.337

Table 9: Teacher-Student Attention Alignment (IoU Scores) for Correct Predictions

Classification

Classification maps observations to discrete labels, including familiar, unknown, hierarchical, or multiple categories.

See all 413 papers

Cross-Dataset Generalization of Bangladeshi Rice Leaf Disease Classifiers: Benchmark, Diagnosis, and Mitigation
Sep 20, 2026Anindya PaulCross-Dataset Generalization Plant Disease Classification
Department of Computer Science and Engineering BRAC University, Dhaka, Bangladesh
BananaVLM: A Domain-Adapted Vision Language Model for Banana Crop Disease Diagnosis
Aug 27, 2026Sangam Kumar Jena, Pandarasamy ArjunanPlant Disease Classification VLM Adaptation
Department of Cyber-Physical Systems Indian Institute of Science Bangalore 560 012, India
DenseSwinV2: Channel Attentive Dual Branch CNN Transformer Learning for Cassava Leaf Disease Classification
Mar 26, 2026Shah Saood, Saddam Hussain KhanPlant Disease Classification Hybrid CNN-Transformer Architectures
Artificial Intelligence Lab, Department of Computer Systems Engineering, University of Engineering and Applied Sciences · Artificial Intelligence Lab, Department of Computer Systems Engineering, University of Engineering and Applied Sciences (UEAS), Swat 19060, Pakistan · Interdisciplinary Research Center for Smart Mobility and Logistics +1

Sep 20, 2026J/K move · Enter open · S save

Cross-Dataset Generalization of Bangladeshi Rice Leaf Disease Classifiers: Benchmark, Diagnosis, and Mitigation

Anindya Paul

Department of Computer Science and Engineering BRAC University, Dhaka, Bangladesh

Cross-Dataset Generalization Plant Disease Classification

Open paper AskPDF arXiv

Cross-dataset transfer in rice leaf disease classification remains a significant challenge, with models trained on one image collection performing substantially worse when deployed on another. We conduct a systematic benchmark across three Bangladeshi rice leaf disease datasets (5,419 images, 6 transfer pairs, 3 CNN backbones, 3 random seeds) to characterize and diagnose this failure. Strong augmentation recovers a mean cross-dataset macro-F1 improvement of +0.070 (Wilcoxon p < 0.001, 15 of 18 transfer pairs positive). Removing non-leaf image content via segmentation shows directional benefit (mean +0.066, p = 0.062, n = 36 paired observations) that is consistent across two independent segmentation methods but does not reach conventional significance. A self-supervised ViT control (DINOv2 linear probe) exhibits equivalent cross-dataset collapse to CNNs, ruling out architecture inductive bias as the primary driver and pointing to acquisition-condition shift. Adaptive batch normalization uniformly harms transfer performance, with harm magnitude correlating with source-target label-prior divergence and model depth (Spearman rho = 0.621, p = 0.009). Grad-CAM attribution analysis on 12 sampled predictions does not distinguish correct from incorrect cross-domain predictions (p = 0.462), indicating that common attribution proxies are insufficient for diagnosing shift at practical sample sizes. We document all frozen results, prespecified analysis criteria, and reproducibility artifacts in a public repository with SHA-256 integrity verification. This work establishes a rigorous empirical baseline for understanding cross-dataset generalization in agricultural computer vision and identifies both effective (augmentation) and ineffective (AdaBN) adaptation strategies.

Abstract

Classification

Explore similar work

Cross-Dataset Generalization of Bangladeshi Rice Leaf Disease Classifiers: Benchmark, Diagnosis, and Mitigation

BananaVLM: A Domain-Adapted Vision Language Model for Banana Crop Disease Diagnosis

DenseSwinV2: Channel Attentive Dual Branch CNN Transformer Learning for Cassava Leaf Disease Classification