大豆室内考种图像数据集  

A dataset of images for indoor soybean survey

作　　者：张哲 饶元[1,2,3] 束雅莉 陈浩然 朱尚尚 王晓波 金秀[1,2,3] 王丰仪[1,2,3] 李佳佳 徐文强 吴康磊[1,2,3] 王安然 ZHANG Zhe;RAO Yuan;SHU Yali;CHEN Haoran;ZHU Shangshang;WANG Xiaobo;JIN Xiu;WANG Fengyi;LI Jiajia;XU Wenqiang;WU Kanglei;WANG Anran(College of Information and Artificial Intelligence,Anhui Agricultural University,Hefei 230036,P.R.China;Key Laboratory of Agricultural Sensors,Ministry of Agriculture and Rural Affairs,Hefei 230036,P.R.China;Anhui Provincial Key Laboratory of Smart Agricultural Technology and Equipment,Hefei 230036,P.R.China;College of Agronomy,Anhui Agricultural University,Hefei 230036,P.R.China)

机构地区：[1]安徽农业大学信息与人工智能学院,合肥230036 [2]农业农村部农业传感器重点实验室,合肥230036 [3]智慧农业技术与装备安徽省重点实验室,合肥230036 [4]安徽农业大学农学院,合肥230036

出　　处：《中国科学数据(中英文网络版)》2025年第1期112-127,共16页China Scientific Data

基　　金：国家自然科学基金(32371993);安徽省重点研究和开发计划面上攻关项目(2023n06020057,202204c06020022);安徽省高校自然科学研究重大项目(2022AH040125)。

摘　　要：选育优质高产品种是提升大豆品质和产量的重要保障。传统大豆考种方法存在费时费力、数据误差大等问题,距大豆表型数据的一站式高通量获取尚有不小的差距。因此,构建大豆室内考种图像数据集,对于开展大豆植株表型信息的高通量准确获取方法研究,实现大豆的自动智能考种尤为重要。本数据集包含原始图像数据、标注文件数据、测试图像数据三部分,涵盖了典型大豆品种的离体豆荚和植株,其中离体豆荚包括无重叠离体豆荚和重叠离体豆荚,大豆植株包括单分枝型、双分枝型和复杂分枝型;标注文件数据包含了大豆植株主茎的实例分割标注、主茎节的检测框标注、在体豆荚的检测框标注、离体豆荚的实例分割和检测框标注,其中在体豆荚和离体豆荚的检测框标注采用了正框和旋转框两种标注方式,共计4.2GB。通过本数据集进行训练和验证的模型在无重叠离体豆荚、重叠豆荚以及大豆植株形态等各项指标上均表现出了良好的检测和分割效果。本数据集可为大豆植株的离体豆荚检测、在体豆荚检测与植株形态分析的目标检测和实例分割模型方法研究提供宝贵的基础图像数据资源,并且对于促进大豆关键考种信息的一站式高通量获取等智能化考种方法研究也具有重要价值。Breeding superior varieties is an important guarantee for enhancing the quality and yield of soybeans.However,traditional soybean phenotyping methods are often time-consuming,labor-intensive,and prone to data,which is a far cry from the one-stop high-throughput acquisition of soybean phenotypic information.Therefore,building a dataset of images for indoor soybean survey is very important for developing methods for high-throughput and accurate phenotypic data acquisition and for further paving the way for automated and intelligent soybean survey.This dataset consists of three parts:original image data,annotation file data,and test image data.It features in-vitro pods and plants of typical soybean varieties,in which the former includes non-overlapping in-vitro pods and overlapping ones,and the latter includes plants of single branch,double branches and complex branches.The annotation file data includes instance segmentation labeling of main stems in soybean plants,detection box labeling of main stem nodes,detection box labeling of in-body pods,and the labeling of instance segmentation and detection box of the in-vitro pods.Moreover,the horizontal and rotation boxes are employed regarding the detection box labeling of soybean plant pods,and in-vitro pods,with a total amount of 4.2 GB.The models trained and validated on this dataset have demonstrated excellent detection and segmentation performance across various metrics,effectively handling non-overlapping and overlapping in-vitro pods,as well as soybean plant structures.This dataset can provide foundational image data for in-vitro pod detection,in-body pod detection and plant morphological analysis of object detection and instance segmentation models.Furthermore,it is of great value for advancing the research on intelligent soybean survey for the one-stop high-throughput acquisition of soybean key information.

关 键 词：大豆 智能考种 目标检测 实例分割 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] S565.1[自动化与计算机技术—计算机科学与技术]