Identification of maize seed vigor under different accelerated aging times using hyperspectral imaging and spectral deep features

作     者：Zhu, Hongfei Yang, Ranbing Lu, Miaomiao Shi, Weiming Sun, Wenbin Lv, Danyang Liu, Hang Wu, Qiong Jiang, Xuwen Han, Zhongzhi 

作者机构：Hainan Univ Haikou 570100 Peoples R China Qingdao Agr Univ Qingdao 266109 Peoples R China Minist Agr & Rural Affairs Key Lab Trop Intelligent Agr Equipment Sanya 572025 Peoples R China 

出 版 物：《COMPUTERS AND ELECTRONICS IN AGRICULTURE》 (Comput. Electron. Agric.)

年 卷 期：2025年第231卷

核心收录：

学科分类：09[农学] 0901[农学-作物学] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

基　　金：National Natural Science Foundation of China [32272003, 52265029] China National Key R & D Program during the 14th Five-year Plan Period [2023YFD2000400] Graduate Innovation Research Project in Hainan Province [Qhyb2024-04] Taishan Scholars Project of Shandong Province [2021-216] Natural Science Foundation of Shandong Province [ZR2021MC107] Shennong Youth Talent Program Major Science and Technology Innovation Project of Shandong Province [2021TZXD003-003, 2021LZGC026-03, 2021LZGC026-05, 2023LZGC008-001] Science and Technology smes Promotion Project of Shandong Province [2021TSGC1016, 2022TSGC1114] Central Government Guide local development Fund [22134-zyyd-nsh, 23139-zyyd-nsh] National Key Research and Development Program [2022YFD2300101-1] 

主　　题：Maize Seed Vigor Spectral Reconstruction Deep Features Ensemble Learning 

摘      要：Maize seed vigor plays a crucial role in improving crop yield. This study proposes a novel method for detecting maize seed vigor via the MAC-seq2seq model. The model successfully reconstructs the spectral sequence of maize seeds. After optimization and iteration, the model achieved a root mean square error (RMSE) of 2.818 x 10-6 on the test set. Subsequently, deep spectral features were extracted using the pretrained MAC-seq2seq, and an ensemble learning model was applied to detect the seed aging hours, with an accuracy of 93.33 %. Additionally, by labeling the spectral sequences with seed vigor index, the combination of ensemble learning and deep spectral features achieved a detection accuracy of 97.62 % for seed vigor. Ultimately, an online seed vigor detection platform was constructed, which effectively measures seed aging time and vigor, was constructed. This study provides an efficient strategy for the nondestructive detection of maize seed vigor, significantly enhancing detection accuracy.