Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

WalnutData: A UAV Remote Sensing Dataset of Green Walnuts and Model Evaluation

Mingjie Wu, Chenggui Yang, Huihua Wang, Chen Xue, Yibo Wang, Haoyu Wang, Yansong Wang, Can Peng, Yuqi Han, Ruoyu Li, Lijun Yun, Zaiqing Chen, Yuelong Xia

TL;DR

This paper introduces WalnutData, the first large-scale, low-altitude UAV dataset for green walnut detection that captures four environmental states reflecting lighting and occlusion (A1, A2, B1, B2). The dataset aggregates 30,240 RGB images across 8 plots in Yunnan, with 706,208 annotated instances, and provides bounding-box annotations in VOC, COCO, and YOLO formats. It offers a thorough benchmark study of major one-stage and two-stage detectors (via Ultralytics and MMDetection), establishing baseline performance and insights on object size and lighting conditions under UAV imagery. The work aims to advance automated harvesting, precision management, and broader agricultural CV research by providing a rich, publicly available resource and rigorous evaluation protocols.

Abstract

The UAV technology is gradually maturing and can provide extremely powerful support for smart agriculture and precise monitoring. Currently, there is no dataset related to green walnuts in the field of agricultural computer vision. Thus, in order to promote the algorithm design in the field of agricultural computer vision, we used UAV to collect remote-sensing data from 8 walnut sample plots. Considering that green walnuts are subject to various lighting conditions and occlusion, we constructed a large-scale dataset with a higher-granularity of target features - WalnutData. This dataset contains a total of 30,240 images and 706,208 instances, and there are 4 target categories: being illuminated by frontal light and unoccluded (A1), being backlit and unoccluded (A2), being illuminated by frontal light and occluded (B1), and being backlit and occluded (B2). Subsequently, we evaluated many mainstream algorithms on WalnutData and used these evaluation results as the baseline standard. The dataset and all evaluation results can be obtained at https://github.com/1wuming/WalnutData.

WalnutData: A UAV Remote Sensing Dataset of Green Walnuts and Model Evaluation

TL;DR

This paper introduces WalnutData, the first large-scale, low-altitude UAV dataset for green walnut detection that captures four environmental states reflecting lighting and occlusion (A1, A2, B1, B2). The dataset aggregates 30,240 RGB images across 8 plots in Yunnan, with 706,208 annotated instances, and provides bounding-box annotations in VOC, COCO, and YOLO formats. It offers a thorough benchmark study of major one-stage and two-stage detectors (via Ultralytics and MMDetection), establishing baseline performance and insights on object size and lighting conditions under UAV imagery. The work aims to advance automated harvesting, precision management, and broader agricultural CV research by providing a rich, publicly available resource and rigorous evaluation protocols.

Abstract

The UAV technology is gradually maturing and can provide extremely powerful support for smart agriculture and precise monitoring. Currently, there is no dataset related to green walnuts in the field of agricultural computer vision. Thus, in order to promote the algorithm design in the field of agricultural computer vision, we used UAV to collect remote-sensing data from 8 walnut sample plots. Considering that green walnuts are subject to various lighting conditions and occlusion, we constructed a large-scale dataset with a higher-granularity of target features - WalnutData. This dataset contains a total of 30,240 images and 706,208 instances, and there are 4 target categories: being illuminated by frontal light and unoccluded (A1), being backlit and unoccluded (A2), being illuminated by frontal light and occluded (B1), and being backlit and occluded (B2). Subsequently, we evaluated many mainstream algorithms on WalnutData and used these evaluation results as the baseline standard. The dataset and all evaluation results can be obtained at https://github.com/1wuming/WalnutData.

Paper Structure

This paper contains 14 sections, 3 figures, 10 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Annotated Image Datasets Collected by UAV
  4. Annotated Datasets in Agriculture
  5. WalnutData Construction
  6. Data Collection
  7. Dataset Construction
  8. Data Annotation
  9. Dataset Split
  10. Dataset Analysis
  11. Experimental Evaluation
  12. Baselines of One-Stage Object Detection Algorithms
  13. Baselines of two-stage object detection algorithms
  14. Conclusion

Figures (3)

  • Figure 1: Examples of local image categories in the WalnutData. Category A1 represents green walnuts that are illuminated by frontal light and unoccluded. Category B1 represents green walnuts that are illuminated by frontal light and occluded. Category B2 represents green walnuts that are backlit and obstructed. Category A2 represents green walnuts that are backlit and unobstructed.
  • Figure 2: The proportion information of the number of instances in each category after the dataset is partitioned. The proportions of the numbers of A1, B1, B2, and A2 instances are similar in the Train, the Val, and the Test respectively.
  • Figure 3: The distribution of the average grayscale values of each instance in the Train, the Val, and the Test. (a), (b), and (c) are the statistics of the grayscale values of each instance in the Train, the Val, and the Test respectively. Among them, 76.31%, 75.59%, and 75.81% of the instances in the Train, the Val, and the Test respectively have grayscale values lower than the median grayscale value (127.5), indicating that more than half of the green walnuts in WalnutData receive less light.