SPP-extractor: Automatic phenotype extraction for densely grown soybean plants
Wan Zhou,
Yijie Chen,
Weihao Li,
Cong Zhang,
Yajun Xiong,
Wei Zhan,
Lan Huang,
Jun Wang,
Lijuan Qiu
Affiliations
Wan Zhou
College of Computer Science, Yangtze University, Jingzhou 434023, Hubei, China
Yijie Chen
College of Agriculture, Yangtze University, Jingzhou 434025, Hubei, China; MARA Key Laboratory of Sustainable Crop Production in the Middle Reaches of the Yangtze River (Co-construction by Ministry and Province), College of Agriculture, Yangtze University, Jingzhou 434025, Hubei, China
Weihao Li
College of Computer Science, Yangtze University, Jingzhou 434023, Hubei, China
Cong Zhang
College of Computer Science, Yangtze University, Jingzhou 434023, Hubei, China
Yajun Xiong
College of Agriculture, Yangtze University, Jingzhou 434025, Hubei, China
Wei Zhan
College of Computer Science, Yangtze University, Jingzhou 434023, Hubei, China
Lan Huang
College of Computer Science, Yangtze University, Jingzhou 434023, Hubei, China; Corresponding authors.
Jun Wang
College of Agriculture, Yangtze University, Jingzhou 434025, Hubei, China; MARA Key Laboratory of Sustainable Crop Production in the Middle Reaches of the Yangtze River (Co-construction by Ministry and Province), College of Agriculture, Yangtze University, Jingzhou 434025, Hubei, China; Corresponding authors.
Lijuan Qiu
National Key Facility for Gene Resources and Genetic Improvement/Institute of Crop Sciences, Chinese Academy of Agricultural Sciences, Beijing 100081, China; Corresponding authors.
Automatic collecting of phenotypic information from plants has become a trend in breeding and smart agriculture. Targeting mature soybean plants at the harvesting stage, which are dense and overlapping, we have proposed the SPP-extractor (soybean plant phenotype extractor) algorithm to acquire phenotypic traits. First, to address the mutual occultation of pods, we augmented the standard YOLOv5s model for target detection with an additional attention mechanism. The resulting model could accurately identify pods and stems and could count the entire pod set of a plant in a single scan. Second, considering that mature branches are usually bent and covered with pods, we designed a branch recognition and measurement module combining image processing, target detection, semantic segmentation, and heuristic search. Experimental results on real plants showed that SPP-extractor achieved respective R2 scores of 0.93–0.99 for four phenotypic traits, based on regression on manual measurements.
