杨梅树识别深度学习模型实验数据集

杨梅树[Myrica rubra (Lour.) S. et Zucc]是一种常绿乔木，树高5-15 m，胸径可达60 cm，树冠可达5 m以上。杨梅树广泛分布在我国江南地区，生长于海拔1500 m以下酸性红壤、山坡向阳的地理环境中。杨梅果是具有江南地理特色的水果。作者于2019年1月23日-24日选择浙江省永嘉县大洋山森林公园作为试验区，采用大疆Phantom4无人机进行航拍，在此基础上对杨梅树树冠进行多边形标记，即采用Mask RCNN（Region Convolutional Neural Networks）深度学习模型对杨梅树进行自动识别，对识别结果运用目视解译方法验证。结果表明，Mask RCNN在杨梅树识别方面有较高精度，总体检出率达90.08%，错检率为9.62%，漏检率为9.92%。杨梅树识别深度学习模型实验数据集包括：（1）杨梅树实验样区(浙江省永嘉县大洋山森林公园) 照片3080张，每张照片像素尺寸为5472 x 3648；（2）杨梅树树冠样本标记数据（298张）；（3）杨梅树深度学习模型识别结果数据。该数据集以.jpg、.json格式存储，由3690个数据文件组成，数据量为25.6 GB（压缩为71个文件，25.5 GB）。

*Myrica rubra* (Lour.) S. et Zucc., commonly known as Chinese bayberry, is an evergreen arbor with a height of 5–15 m, a diameter at breast height (DBH) of up to 60 cm, and a crown width exceeding 5 m. It is widely distributed in the Jiangnan region of China, thriving in acidic red soils at elevations below 1500 m on sun-facing mountain slopes. The fruit of Chinese bayberry is a specialty agricultural product unique to the Jiangnan region.

On January 23–24, 2019, the authors selected Dayangshan Forest Park in Yongjia County, Zhejiang Province as the experimental area, and conducted aerial photography using a DJI Phantom 4 unmanned aerial vehicle (UAV). Polygonal annotations were performed for the crowns of the captured Chinese bayberry trees, and the Mask R-CNN (Region Convolutional Neural Networks) deep learning model was adopted for automatic identification of the trees. The identification results were then verified via visual interpretation.

The experimental results showed that Mask R-CNN exhibited high accuracy in Chinese bayberry tree identification, with an overall detection rate of 90.08%, a false detection rate of 9.62%, and a missed detection rate of 9.92%.

The experimental dataset for Chinese bayberry tree identification via deep learning includes three components: (1) 3080 photos of the experimental sample area (Dayangshan Forest Park, Yongjia County, Zhejiang Province), with each photo having a pixel size of 5472 × 3648; (2) 298 crown annotation samples of Chinese bayberry trees; (3) identification result data of the Chinese bayberry deep learning model.

This dataset is stored in .jpg and .json formats, consisting of 3690 individual data files, with a total size of 25.6 GB (compressed into 71 files with a total size of 25.5 GB).

搜集汇总

背景与挑战

背景概述

该数据集包含3080张无人机拍摄的杨梅树影像和标记样本，用于训练和验证Mask RCNN深度学习模型，识别准确率达90.08%。数据总大小为25.6 GB，格式为.jpg和.json。