Vision normalizing flows for the probability-informed detection of banana diseases from in-field images
- 1Independent Researcher
- 2Epidemiology and Modelling Group, Department of Plant Sciences, University of Cambridge
Abstract
Introduction: Banana diseases impose severe production losses in tropical smallholder farming systems, yet accurate in-field visual diagnosis remains difficult because symptom expression varies across cultivars and growth stages and several diseases produce morphologically overlapping foliar signs. We therefore developed a probabilistic image-recognition framework for detecting five economically important banana diseases from in-field photographs without disease-specific fine-tuning of the vision backbone.
Methods: Frozen 1,152-dimensional embeddings were extracted from the DINOv3 vision foundation model and coupled with a conditional normalizing flow. The model was trained on four publicly available datasets comprising approximately 95,000 images across eight classes spanning diseased banana plants, healthy tissue, non-banana vegetation, and general natural imagery.
Results: On an independent test set, the model achieved F1 scores exceeding 0.98, average precision values of 0.9678-0.9998, and AUROC values of 0.9967-0.9998 across all five diseases evaluated as binary detection problems. Multi-class accuracy was nearperfect, with limited confusion between Yellow Sigatoka and Black Sigatoka. Because the normalizing flow estimates explicit conditional probability densities rather than decision boundaries, two complementary log-likelihood ratios were derived: a disease ratio comparing each disease class against healthy banana, and a plant ratio comparing banana against non-banana imagery.
Discussion: Together, these likelihood ratios define an interpretable two-dimensional diagnostic space that simultaneously quantifies evidence for disease presence and image relevance, cleanly separating diseased plants, healthy plants, and out-ofdistribution images while flagging uncertain predictions for confirmatory testing. Batched inference costs 6.4 ms per image on a single TPU core, supporting deployment at surveillance-programme scale and providing a scalable, uncertainty-aware diagnostic tool for smallholder farming systems.
Related
- YOLOv9-Based Detection of Diseases in Poplar Trees Using Histogram Equalization and Computer Vision
- ViSSR: Physics-Informed Hybrid Vision-LSTM/State-Space Super-Resolution for Digital Rock Microstructure Analysis
- Comparing pest survey methods through detection probability: seasonal and abundance-dependent detection of spotted lanternfly (Lycorma delicatula, Hemiptera: Fulgoridae)
- Vision In SyndrOmic Neurodevelopmental disorders-15q (VISiON-15q) study: eye tracking to supplement clinical examination for detection of cerebral/cortical visual impairment in children with Angelman and Duplication of 15q Syndromes
- Vision-spectroscopy cooperative sensing and cold atmospheric plasma inactivation for early warning and management of agricultural product diseases
BibTeX
@article{prusokiene2026vision,
title = {{Vision normalizing flows for the probability-informed detection of banana diseases from in-field images}},
author = {Alisa Prusokiene and Augustinas Prusokas and Renata Retkute},
journal = {Frontiers in Plant Science},
year = {2026},
month = oct,
doi = {10.3389/fpls.2026.1937900},
url = {https://www.frontiersin.org/articles/10.3389/fpls.2026.1937900/full},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Vision normalizing flows for the probability-informed detection of banana diseases from in-field images
AU - Alisa Prusokiene
AU - Augustinas Prusokas
AU - Renata Retkute
PY - 2026
DA - 2026/10/09
JO - Frontiers in Plant Science
DO - 10.3389/fpls.2026.1937900
UR - https://www.frontiersin.org/articles/10.3389/fpls.2026.1937900/full
AB - Introduction: Banana diseases impose severe production losses in tropical smallholder farming systems, yet accurate in-field visual diagnosis remains difficult because symptom expression varies across cultivars and growth stages and several diseases produce morphologically overlapping foliar signs. We therefore developed a probabilistic image-recognition framework for detecting five economically important banana diseases from in-field photographs without disease-specific fine-tuning of the vision backbone. Methods: Frozen 1,152-dimensional embeddings were extracted from the DINOv3 vision foundation model and coupled with a conditional normalizing flow. The model was trained on four publicly available datasets comprising approximately 95,000 images across eight classes spanning diseased banana plants, healthy tissue, non-banana vegetation, and general natural imagery. Results: On an independent test set, the model achieved F1 scores exceeding 0.98, average precision values of 0.9678-0.9998, and AUROC values of 0.9967-0.9998 across all five diseases evaluated as binary detection problems. Multi-class accuracy was nearperfect, with limited confusion between Yellow Sigatoka and Black Sigatoka. Because the normalizing flow estimates explicit conditional probability densities rather than decision boundaries, two complementary log-likelihood ratios were derived: a disease ratio comparing each disease class against healthy banana, and a plant ratio comparing banana against non-banana imagery. Discussion: Together, these likelihood ratios define an interpretable two-dimensional diagnostic space that simultaneously quantifies evidence for disease presence and image relevance, cleanly separating diseased plants, healthy plants, and out-ofdistribution images while flagging uncertain predictions for confirmatory testing. Batched inference costs 6.4 ms per image on a single TPU core, supporting deployment at surveillance-programme scale and providing a scalable, uncertainty-aware diagnostic tool for smallholder farming systems.
SN - 1664-462X
ER -