南京林业大学汪希伟博士等:基于交替背光与正面光照明及振动辅助的中国典型杨树品种机器视觉高通量种子表型性状测量

High-throughput seed phenotyping of Populus cultivars in China using vibration-assisted machine vision with alternating back-lit and front-lit illuminations

Xiwei Wang^1*, Mubikayi Muhong Horly¹, Zanpeng Li¹, Maocheng Zhao¹, Bin Wu¹, Mengmeng Wang¹, Jiale Deng¹, Mengmeng Qiao¹, Xiao Chen¹, Yongjie Lu¹, Liang Qi¹, Weijun Xie¹, Hongyuan Zou¹, Yan Zhang², Rusheng Peng²

(1.College of Mechanical and Electronic Engineering, Nanjing Forestry University, Nanjing 210037, China; 

2. Liaoning Institute of Poplar Research, Gaizhou 115213, Liaoning, China)

Abstract: Seeds of major Populus cultivars were collected from across China in 2024 to build the image-data bank of over 1 187 000 images of singular seeds for the National Forestry and Grassland Science Data Center (NFGSDC). An innovative vibration assisted machine-vision system was built with alternating back-lit and front-lit illumination, which incorporated a flexible vibratory panel (FVP) to manipulate the multitude of seeds to minimize the occurrence of butting or overlapping, and the lighting from alternating directions to capture phenotypic features both in silhouettes and in vivid color images. To investigate how illumination directions would affect phenotyping, morphological and chromatic metrics were measured, respectively from only the common front-lit images and through the combined use with back-lit images, and applied to distinguish different cultivars and harvest-batches. Results verified that back-lit excelled for reliable segmentation for feature images and accurate morphological metrics, especially when the closeness was clearly revealed in the clustering dendrogram between Nanlin 895 and Zhonglin 46, which shared a common genetic sourcing from P. Euramericana. In contrast, front-lit images were prone to occasional segmentation defects leading to inaccurate morphological measurements due to the highly dynamic range of seed colors, which caused the clustering to lose the genetic relevance. The power of the image-dataset of alternating illuminations was further demonstrated when a decent accuracy of 0.819 yielded from the simple support-vector-machine classification while working on only the back-lit morphological measurements, and the increase to 0.856 with statistical significance if with the addition of chromatic metrics from corresponding front-lit color images, while other image characteristics had been strictly held back. The vibration-assisted alternating illumination protocol established in this work to capture delicate seed-features of Populus cultivars may also be applied to other small grains facing similar imaging challenges, laying a sturdy step-stone of high throughput phenotyping for large-scale breeding programs and genetic studies.

Keywords:  Populus seed, machine vision, camera calibration, flexible vibratory plate, back-lit and front-lit illumination

DOI: 10.25165/j.ijabe.20261901.9850

Citation:Wang X W, Horly M M, Li Z P, Zhao M C, Wu B, Wang M M, et al. High-throughput seed phenotyping of Populus cultivars in China using vibration-assisted machine vision with alternating back-lit and front-lit illuminations. Int J Agric & Biol Eng, 2026; 19(1): 197–212.