Artificial neural networks discriminate lettuce seeds with different levels of thermoinhibition

Abstract: The thermoinhibition of lettuce seed germination causes important losses for producers, who do not have thermotolerant commercial cultivars. One of the obstacles has been the scarcity of optimizing techniques capable of efficiently discriminating thermotolerant and thermosensitive cultivars. The aim of this work was to evaluate the use of neural networks to discriminate different levels of thermoinhibition in lettuce seeds. Seeds of 18 cultivars were evaluated for thermoinhibition considering the characteristics of the first and last germination count and germination speed index, in seeds subjected to temperatures of 20, 25, 30 and 35 °C. The remaining seeds, which did not germinate, were subjected to the tetrazolium test. Analyses were performed immediately after seed harvesting and repeated after six months of storage. Discriminant analysis was performed and the Kohonen’s Self-Organizing Map (SOM) was created using Artificial Neural Networks (ANNs). Neural networks discriminate lettuce cultivars and organizes them in terms of seed thermoinhibition tolerance through Kohonen’s Self-Organizing Map. Discriminant analysis consistently identifies the Everglades and Luiza genotypes as tolerant to thermoinhibition.

摘要：生菜种子萌发热抑制现象给缺乏耐热商用品种的种植者造成了重大经济损失。此前制约该领域发展的一大瓶颈在于，缺乏能够高效区分耐热与热敏生菜品种的优化技术方案。
本研究旨在评估人工神经网络（Artificial Neural Networks, ANNs）在区分生菜种子不同程度热抑制响应中的应用效果。研究选取18个生菜品种的种子，在20℃、25℃、30℃及35℃恒温条件下，以首次萌发计数、末次萌发计数及萌发速率指数为评价指标开展热抑制特性评估；对未萌发的剩余种子，采用四唑（tetrazolium）试验进行活力检测。所有检测均在种子收获后即刻开展，并于贮藏6个月后重复进行一次。
本研究采用判别分析法，并基于人工神经网络构建了科赫内自组织映射（Kohonen’s Self-Organizing Map, SOM）模型。结果表明，基于科赫内自组织映射的人工神经网络可有效区分生菜品种，并依据种子热抑制耐受能力对其进行聚类分级；判别分析法可稳定鉴定出Everglades与Luiza两个基因型生菜具有热抑制耐受性。