AppleScab-LT: A Longitudinal Real-Field Apple Scab Dataset for Temporal Disease Progression Analysis

Aamir Hilal; Shabir Ahmad Sofi; Neeraj Goel

cs.CVAug 14, 2026

AppleScab-LT: A Longitudinal Real-Field Apple Scab Dataset for Temporal Disease Progression Analysis

Authors: Aamir Hilal, Shabir Ahmad Sofi, Neeraj Goel

Organizations: Department of Information Technology, National Institute of Technology, Hazratbal Srinagar 190006, Jammu & Kashmir, India · Department of Computer Science and Engineering, Indian Institute of Technology, Ropar Rupnagar 140001, Punjab, India

Abstract

The development of reliable plant disease monitoring systems is constrained by limited longitudinal datasets capturing disease progression under natural field conditions. Although existing plant disease datasets have advanced image-based recognition, most consist of static images acquired at a single time point, limiting analysis of temporal disease evolution and severity progression. To address this gap, this study presents AppleScab-LT, a longitudinal real-field dataset developed to monitor apple scab progression through repeated observations of individually tracked infected leaves. Guided by a research-question-driven framework, the dataset was systematically developed, validated, and characterized for reliable longitudinal disease analysis. AppleScab-LT was constructed through systematic orchard monitoring under natural environmental conditions, incorporating longitudinal leaf tracking, expert-guided disease verification, polygon-based annotation, leaf isolation, disease severity quantification, and temporal sequence construction. A comprehensive quality assurance framework, including standardized annotation protocols, expert validation, automated integrity checks, sequence-level verification, and temporal consistency analysis, was applied throughout curation. The dataset contains 21 longitudinal leaf sequences, 2,101 high-resolution images, and 264 progressive temporal samples from repeated monitoring of same infected leaves. It captures variability in severity accumulation, progression rates, monitoring duration, and inter-leaf progression. Quantitative disease descriptors based on pixel severity, color-intensity severity, and normalized relative severity provide standardized measurements for temporal disease analysis. AppleScab-LT provides a reliable resource for temporal disease intelligence, disease progression modelling, precision agriculture, and future crop health monitoring

Figures & tables

Dataset	Crop	Field	Temporal	Longitudinal	Segmentation	Expert Verified	Disease Severity	Repeated Monitoring
Single-Plant Datasets
ALDD Apple Dataset [Jiang et al., 2019] [ 10 ]	1	No	No	No	No	Yes	No	No
Soybean Dataset [Wu et al., 2019] [ 11 ]	1	No	No	No	No	Yes	No	No
Citrus Dataset [Dananjayan et al., 2022] [ 12 ]	1	No	No	No	No	Yes	No	No
Rice Dataset [Joshi et al., 2022] [ 13 ]	1	No	No	No	No	Yes	No	No
Apple Plant Dataset [Khan et al., 2022] [ 14 ]	1	No	No	No	No	Yes	No	No

Table 1: Comparison of representative plant disease datasets with respect to field acquisition, temporal monitoring, annotation characteristics, and progression information.

Dataset	Field	Longitudinal	Repeated Leaf Tracking	Severity Labels	Segmentation	Temporal Metadata	Progression Analysis
Single-Plant Apple Datasets
ALDD Apple Dataset [ 10 ]	$\times$	$\times$	$\times$	$\times$	$\times$	$\times$	$\times$
Apple Plant Dataset [ 14 ]	$\times$	$\times$	$\times$	$\times$	$\times$	$\times$	$\times$
Plant Pathology Dataset [ 8 ]	✓	$\times$	$\times$	$\times$	$\times$	$\times$	$\times$
ADLD (Apple Leaf Disease) [ 36 ]	✓	$\times$	$\times$	$\times$	✓	$\times$	$\times$
Multi-Plant Datasets (containing Apple)

Table 2: Comparison of representative apple disease datasets.

Category	Dataset	Crop	Field	Repeated	Longitudinal	Temporal	Manual	Metadata &	Expert	Leaf-level	Disease
				Monitoring		Severity Labels	Polygon Annotation	Temporal Descriptors	Disease Verification	Tracking	Progression Analysis
General Plant Datasets	PlantVillage [ 6 ]	Multiple	$\times$	$\times$	$\times$	$\times$	$\times$	$\times$	$\checkmark$	$\times$	$\times$
	PlantDoc [ 7 ]	Multiple	$\checkmark$	$\times$	$\times$	$\times$	$\times$	$\times$	$\checkmark$	$\times$	$\times$
	PDD271 [ 42 ]	Multiple	$\checkmark$	$\times$	$\times$	$\times$	$\times$	$\times$	$\checkmark$	$\times$	$\times$
	PlantSeg [ 47 ]	Multiple	$\checkmark$	$\times$	$\times$	$\times$	$\checkmark$	$\times$	$\checkmark$	$\times$	$\times$
	ADLD [ 36 ]	Apple	$\checkmark$	$\times$	$\times$	$\times$	$\checkmark$	$\times$	$\checkmark$	$\times$	$\times$

Table 3: Taxonomy and comparison of representative general and longitudinal plant disease datasets.

Figure 1: Geographical locations of the two study locations used for constructing the AppleScab-LT dataset.

Characteristic	Study Site 1	Study Site 2
Location	Sher-e-Kashmir University of Agricultural Sciences and Technology of Kashmir (SKUAST-K), Shalimar, Srinagar	Commercial Apple Orchard, Shopian District
Type	Research Orchard	Commercial Orchard
Region	Srinagar, Jammu and Kashmir, India	Shopian, Jammu and Kashmir, India
Monitoring Period	April 2025 – September 2025	April 2025 – September 2025
Climate	Temperate Himalayan Climate	Temperate Himalayan Climate
Disease	Apple Scab ( Venturia inaequalis )	Apple Scab ( Venturia inaequalis )

Table 4: Characteristics of the study sites selected for longitudinal data collection.

Parameter	Specification
Camera	Canon EOS R50
Sensor Type	APS-C CMOS
Image Resolution	24.2 Megapixels
Lens	Canon RF 24–105 mm F4–7.1 IS STM
Sensor Resolution	24.2 MP APS-C CMOS
Image Format	RGB (JPEG)

Table 5: Specifications of the image acquisition system used during field monitoring.

Figure 2: Longitudinal monitoring protocol adopted for temporal dataset construction.

Figure 3: Overview of the multi-disease apple repository developed during the field campaign.

Figure 4: Representative longitudinal apple scab sequence from leaf LS12, LS06, LS05 and LS03 illustrating disease progression across multiple observation days.

Figure 5: Construction pipeline of the AppleScab-LT dataset.

Curation Stage	Count
Original multi-disease repository	55,849 images
Apple scab images (single + multi-disease)	7,497 images
Candidate repeated observations	2,964 images
Verified leaf sequences	21 sequences
Images retained after quality assessment	2,101 images
Expert verified images	2,101 images

Table 6: Quantitative summary of the AppleScab-LT dataset curation process.

Figure 6: AppleScab-LT dataset curation pipeline.

Figure 7: Annotation workflow adopted for AppleScab-LT.

Figure 8: Representative examples illustrating the standardized annotation protocol adopted for AppleScab-LT under varying levels of annotation complexity. Each example presents the original RGB image, polygon annotations, generated leaf mask, and disease mask. (a) Simple Annotation ,(b) Moderate Annotation Complexity , (c) Challenging Annotation Scenario

Figure 9: Expert disease verification and quality assurance workflow adopted during the construction of the dataset.

Verification Stage	Responsible	Primary Objective	Output	Acceptance Criterion
Primary Annotation	Primary Annotator	Generate polygon annotations for the complete leaf and all visible scab lesions using the standardized annotation protocol.	Initial annotation set	Complete leaf and scab polygons generated.
Independent Review I	Horticultural Expert 1	Verify disease identity, lesion morphology, biological correctness, and polygon accuracy.	Expert review comments	Disease diagnosis and annotation biologically correct.
Independent Review II	Horticultural Expert 2	Independently verify disease diagnosis, temporal consistency, sequence continuity, and annotation completeness.	Independent validation report	Temporal consistency confirmed across repeated observations.
Consensus Review	Primary Annotator + Both Experts	Resolve annotation discrepancies through discussion and refinement whenever required.	Consensus-approved annotation	All identified inconsistencies resolved.
Final Acceptance	Both Experts	Perform final quality assessment before dataset inclusion.	Final approved annotation	$\checkmark$ Correct disease diagnosis $\checkmark$ Accurate leaf polygon $\checkmark$ Accurate scab polygon $\checkmark$ Temporal consistency verified $\checkmark$ High image quality $\checkmark$ Complete metadata $\checkmark$ Approved by both experts

Table 7: Verification and acceptance protocol adopted during the construction of the AppleScab-LT dataset.

Figure 10: Technical quality assurance framework adopted during the construction of AppleScab-LT

Figure 11: Proposed reliability validation framework for AppleScab-LT.

Reliability Dimension	Validation Method	Purpose
Spatial Annotation Reliability	Polygon annotation protocol, expert review, automated polygon validation	Ensure accurate delineation of leaf and scab regions.
Expert Verification Reliability	Independent review by two horticultural experts and consensus-based refinement	Ensure biological correctness of disease annotations.
Temporal Sequence Reliability	Sequence verification, chronological ordering, identity preservation	Ensure continuity of repeated observations of the same leaf.
Dataset Integrity Reliability	Automated Python validation scripts and manual inspection	Ensure structural consistency, metadata integrity, and dataset completeness.

Table 8: Reliability validation dimensions adopted for AppleScab-LT.

Validation Aspect	Verification Procedure	Outcome
Annotation Quality	Standardized polygon annotation and expert review	Verified
Disease Identity	Independent review by two horticultural experts	Verified
Temporal Continuity	Leaf sequence verification and chronological validation	Verified
Dataset Integrity	Automated Python validation and manual inspection	Verified
Metadata Consistency	Cross-validation of filenames, sequence IDs, masks, and JSON files	Verified
Dataset Validation	Compliance with predefined acceptance criteria	Verified

Table 9: Summary of reliability validation outcomes for AppleScab-LT.

Category	Attribute	Value
Dataset	Total RGB images	2,101
Dataset	Leaf sequences	21
Annotations	JSON files, Leaf Masks, Disease Masks, Polygon Annotations	2,101
Temporal	Temporal observations	2,101
	Disease stages	4
	Generated temporal samples	264

Table 10: Overall characteristics of the AppleScab-LT longitudinal dataset.

Leaf	Monitoring Duration (Days)	Temporal Observations	Captured Images	Mean Interval (Days)	Images/Observation
LS01	30	12	37	2.73	3.08
LS02	92	9	92	11.50	10.22
LS03	92	28	184	3.41	6.57
LS04	38	17	131	2.38	7.71
LS05	92	33	417	2.88	12.64
LS06	91	30	325	3.14	10.83

Table 11: Leaf-wise longitudinal characteristics of the AppleScab-LT dataset.

Figure 12: Distribution of temporal sequence lengths in the proposed AppleScab-LT dataset.

Figure 13: Disease severity quantification framework adopted in AppleScab-LT.

Figure 14: Longitudinal disease progression dynamics captured by AppleScab-LT.

Figure 15: Inter-leaf variability in normalized disease severity across the AppleScab-LT dataset.

Figure 16: Temporal disease descriptor map of AppleScab-LT generated using ten longitudinal characterization descriptors extracted from each monitored leaf sequence.

Figure 17: Quartile-based disease stage construction employed in AppleScab-LT.

Disease Stage	Severity Range	Samples	Relative Severity Range
Stage I	$\leq Q_{1}$	66	$\leq 0.701$
Stage II	$Q_{1}-Q_{2}$	66	0.7001 – 0.810
Stage III	$Q_{2}-Q_{3}$	66	0.810 – 0.878
Stage IV	$>Q_{3}$	66	$>0.878$
Total	–	264	100	–

Table 12: Distribution of disease stages generated using quartile-based partitioning of normalized relative disease severity.

Figure 18: Representative examples illustrating the visual diversity captured by AppleScab-LT : (a) illumination conditions, (b) background complexity, (c) leaf orientation, and (d) disease severity progression.

Figure 19: Hierarchical organization of the AppleScab-LT dataset.

Research Application	Dataset Support	Recommended Metrics	Research Objective
Disease Classification	RGB images with disease stage labels support binary, multiclass, and progression-aware disease recognition.	Accuracy, Precision, Recall, F1-score, Cohen’s $\kappa$ , ROC-AUC	Evaluate disease recognition performance and stage-wise classification accuracy.
Semantic Segmentation	Polygon annotations together with binary leaf and lesion masks enable precise lesion localization and disease area estimation.	IoU, Dice Coefficient, Precision, Recall	Assess spatial agreement between predicted and reference lesion masks.
Disease Severity Estimation	Pixel severity, colour intensity severity, and normalized relative severity measurements support quantitative disease assessment.	MAE, RMSE, $R^{2}$ , Pearson Correlation	Measure agreement between predicted and reference severity values.
Disease Progression Modelling	Longitudinal severity measurements enable modelling of continuous disease development over time.	MAE, RMSE, $R^{2}$ , Dynamic Time Warping (DTW)	Evaluate continuous disease trajectory prediction throughout disease progression.
Disease Progression Forecasting	Repeated temporal observations facilitate forecasting of future disease growth and stage transitions.	MAE, RMSE, $R^{2}$ , Dynamic Time Warping (DTW)	Assess temporal prediction accuracy and progression forecasting capability.
Longitudinal Disease Tracking	Unique sequence identifiers maintain correspondence of individual leaves throughout the monitoring period.	Track Purity, Identity Preservation Rate, Sequence Consistency	Evaluate temporal consistency and preservation of leaf identity across observations.

Table 13: Research applications supported by AppleScab-LT together with recommended evaluation metrics and corresponding objectives.

Figure 20: Recommended Leave-One-Leaf-Out Cross-Validation (LOLOCV) protocol for AppleScab-LT.

Model	Spatial Encoder	Temporal Module	Accuracy Range	Macro F1 Range
CNN–LSTM	CNN	LSTM	79–84%	0.79–0.84
ViT–LSTM	Vision Transformer	LSTM	81–86%	0.81–0.86

Table 14: Representative reference implementation results on AppleScab-LT.

Agricultural Image Analysis

Agricultural image analysis uses computer vision and multimodal learning to monitor crops, livestock, produce quality, and field conditions.

See all 94 papers

A Dataset-Centric Benchmark of Deep Learning Methods for Grape Leaf Disease Classification and Detection
Aug 20, 2026Petar Canoski, Vlatko Spasev, Ivica Dimitrovski +2Plant Disease Classification Agricultural Image Analysis
Faculty of Computer Science and Engineering, University Ss Cyril and Methodius, ul. Rudzer Boshkovikj 16, P.O. 393, Skopje, 100, North Macedonia
Pixel-Precise Explainable Stress Indexing: A Semantic Segmentation Framework for Disease Severity Quantification in Field Crops
Jul 5, 2026Raunak Kumar, Soumyashree KarAgricultural Image Analysis Image Segmentation
Centre of Studies in Resources Engineering Indian Institute of Technology Bombay Mumbai, Maharashtra, India
SAGE: Scalable Agentic Grounded Evaluation for Crop Disease Diagnosis
May 10, 2026Muhammad Arbab Arshad, Tirtho Roy, Yanben Shen +7VLM Evaluation Agricultural Image Analysis
Iowa State University, USA · New York University, USA

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

A Dataset-Centric Benchmark of Deep Learning Methods for Grape Leaf Disease Classification and Detection

Petar Canoski, Vlatko Spasev, Ivica Dimitrovski, Ivan Kitanovski+1

Faculty of Computer Science and Engineering, University Ss Cyril and Methodius, ul. Rudzer Boshkovikj 16, P.O. 393, Skopje, 100, North Macedonia

Plant Disease Classification Agricultural Image Analysis Cross-Dataset Generalization

Open paper AskPDF arXiv

Grape leaf disease recognition is important for precision agriculture, enabling early diagnosis, timely intervention, and improved vineyard management. Although deep learning has achieved strong results, many studies rely on few datasets, often acquired under controlled conditions, and may not reflect real vineyard challenges such as complex backgrounds, variable illumination, occlusion, leaf pose, disease severity, and device differences. This paper presents a dataset-centric benchmark of deep learning methods for grape leaf disease classification and detection. We analyze publicly available datasets in terms of disease categories, annotation types, acquisition conditions, image characteristics, class distributions, provenance, and task suitability. Representative models are evaluated in three settings: image-level classification, region-level classification, and object detection. Classification is assessed using accuracy, while detection is evaluated using mAP@50 and mAP@50:95. Cross-dataset experiments further examine transfer between datasets with compatible disease categories but different visual and annotation characteristics. Results show near-saturated classification performance on several controlled or derivative datasets, greater difficulty on heterogeneous datasets, and substantial variation in detection performance across annotation settings. Cross-dataset performance drops sharply, especially for object detection, indicating that shared disease labels do not necessarily define equivalent recognition tasks. The benchmark emphasizes dataset provenance, realistic field evaluation, annotation compatibility, and external validation for reliable vineyard disease recognition.

Abstract

Agricultural Image Analysis

Explore similar work

A Dataset-Centric Benchmark of Deep Learning Methods for Grape Leaf Disease Classification and Detection

Pixel-Precise Explainable Stress Indexing: A Semantic Segmentation Framework for Disease Severity Quantification in Field Crops

SAGE: Scalable Agentic Grounded Evaluation for Crop Disease Diagnosis