ABSTRACT

Plants are a major source of food all around the world, and they are mainly affected by diseases caused by pathogens, insects, and parasitic plants. If the diseases are identified in the earlier stage, then it will be easy to apply pesticides and prevent the disease from further propagation. To recognise these diseases at earlier stages automatically, different researchers have established artificial intelligence-based approaches using machine learning and deep learning approaches which can identify the diseased lesions with high accuracy. In this paper, a novel Hybrid Moth Flame Optimization Algorithm-butterfly optimisation algorithm (HMFO-BOA) based optimal ensemble deep transfer network (OEDTN) classifier for banana leaf disease identification is developed. The OEDTN architecture offers increased predictability for banana leaf disease using Maximum Mean Discrepancy (MMD), ensemble learning, domain adaptation, and parameter transfer learning. The feature extraction is done by constructing different Deep Transfer Networks (DTN) via diverse kernel MMD. Finally, the DTNs are integrated via ensemble learning to obtain the final classification outcomes. The MFOBOA algorithm assigns optimal voting weights for each DTN to dynamically construct the OEDTN architecture. In the consequent testing process, the input banana leaf disease images are categorised into diverse classes such as BBW (Banana Bacterial Wilt Disease), BBS (Banana Black Sigatoka Disease), Cordana, pestalotiopsis, Sigatoka, and healthy. The experiments conducted on the Banana Leaf and BBW-BBS datasets prove that the proposed model offers improved performance than the state-of-art techniques.

摘要

植物是世界各地的主要食物来源，它们主要受到病原体、昆虫和寄生植物引起的疾病的影响。如果在早期发现病害，那么就很容易施用农药并防止病害进一步传播。为了在早期阶段自动识别这些疾病，不同的研究人员建立了基于人工智能的方法，使用机器学习和深度学习方法，可以高精度识别病变部位。本文开发了一种基于混合蛾火焰优化算法-蝴蝶优化算法（HMFO-BOA）的最优集成深度传输网络（OEDTN）分类器，用于香蕉叶病害识别。OEDTN 架构使用最大平均差异 (MMD)、集成学习、域适应和参数传递学习来提高香蕉叶病的可预测性。特征提取是通过不同的内核MMD构建不同的深度传输网络（DTN）来完成的。最后，通过集成学习对 DTN 进行集成以获得最终的分类结果。MFOBOA算法为每个DTN分配最佳投票权重，动态构建OEDTN架构。在随后的测试过程中，输入的香蕉叶病图像被分为不同的类别，例如BBW（香蕉青枯病）、BBS（香蕉黑叶斑病）、Cordana、鼠疫病、叶斑病和健康。