Yi-Chen Chen , Jen-Cheng Wang , Mu-Hwa Lee , An-Chi Liu , Joe-Air Jiang
{"title":"Enhanced detection of mango leaf diseases in field environments using MSMP-CNN and transfer learning","authors":"Yi-Chen Chen , Jen-Cheng Wang , Mu-Hwa Lee , An-Chi Liu , Joe-Air Jiang","doi":"10.1016/j.compag.2024.109636","DOIUrl":null,"url":null,"abstract":"<div><div>Mango trees affected by various diseases often exhibit distinctive leaf symptoms. Accurate and timely diagnosis is crucial for mango cultivation. Deep learning algorithms provide a viable solution for precisely detection of mango leaf diseases. However, two main challenges exist: environmental interference and the difficulty of collecting leaf image data from the field. To address these challenges, this study introduces a multi-scale and multi-pooling convolutional neural network (MSMP-CNN) model. The proposed model undergoes a pre-training phase, followed by transfer learning and fine-tuning, and ultimately focuses on identifying mango leaf diseases using real-world images. This model exhibits outstanding performance in identifying various mango leaf diseases. The model achieved an accuracy of 95 % on its own. After being enhanced by transfer learning and find-tuning, the model achieved an impressive accuracy of 98.5 %. To compare the classification performance with and without transfer learning and fine-tuning, t-distributed stochastic neighbor embedding (t-SNE) plots were used. Class activation mapping (CAM) heatmaps were also utilized to highlight class-specific regions of images, helping verify whether the model focused on the appropriate parts of the image for disease identification. These findings underscore the strong potential of the model combining with transfer learning and fine-tuning to advance mango leaf disease detection. In the future, the proposed model will evolve into a real-time, precise diagnostic system for mango leaf diseases, thereby transforming mango cultivation management from precision farming to smart agriculture.</div></div>","PeriodicalId":50627,"journal":{"name":"Computers and Electronics in Agriculture","volume":"227 ","pages":"Article 109636"},"PeriodicalIF":7.7000,"publicationDate":"2024-11-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Computers and Electronics in Agriculture","FirstCategoryId":"97","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S0168169924010275","RegionNum":1,"RegionCategory":"农林科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"AGRICULTURE, MULTIDISCIPLINARY","Score":null,"Total":0}
引用次数: 0
Abstract
Mango trees affected by various diseases often exhibit distinctive leaf symptoms. Accurate and timely diagnosis is crucial for mango cultivation. Deep learning algorithms provide a viable solution for precisely detection of mango leaf diseases. However, two main challenges exist: environmental interference and the difficulty of collecting leaf image data from the field. To address these challenges, this study introduces a multi-scale and multi-pooling convolutional neural network (MSMP-CNN) model. The proposed model undergoes a pre-training phase, followed by transfer learning and fine-tuning, and ultimately focuses on identifying mango leaf diseases using real-world images. This model exhibits outstanding performance in identifying various mango leaf diseases. The model achieved an accuracy of 95 % on its own. After being enhanced by transfer learning and find-tuning, the model achieved an impressive accuracy of 98.5 %. To compare the classification performance with and without transfer learning and fine-tuning, t-distributed stochastic neighbor embedding (t-SNE) plots were used. Class activation mapping (CAM) heatmaps were also utilized to highlight class-specific regions of images, helping verify whether the model focused on the appropriate parts of the image for disease identification. These findings underscore the strong potential of the model combining with transfer learning and fine-tuning to advance mango leaf disease detection. In the future, the proposed model will evolve into a real-time, precise diagnostic system for mango leaf diseases, thereby transforming mango cultivation management from precision farming to smart agriculture.
受各种病害影响的芒果树通常会表现出独特的叶片症状。准确及时的诊断对芒果种植至关重要。深度学习算法为精确检测芒果叶片疾病提供了可行的解决方案。然而,目前存在两大挑战:环境干扰和从田间收集叶片图像数据的难度。为应对这些挑战,本研究引入了多尺度和多池化卷积神经网络(MSMP-CNN)模型。该模型经过预训练阶段、迁移学习阶段和微调阶段,最终专注于利用真实世界的图像识别芒果叶病。该模型在识别各种芒果叶病方面表现出色。模型本身的准确率达到 95%。在经过迁移学习和查找调整增强后,该模型的准确率达到了令人印象深刻的 98.5%。为了比较有无迁移学习和微调的分类性能,使用了 t 分布随机邻域嵌入(t-SNE)图。此外,还使用了类激活图谱(CAM)热图来突出图像的特定类区域,以帮助验证模型是否侧重于图像的适当部分进行疾病识别。这些发现凸显了该模型与迁移学习和微调相结合在推进芒果叶病害检测方面的强大潜力。未来,该模型将发展成为一个实时、精确的芒果叶病诊断系统,从而将芒果种植管理从精准农业转变为智慧农业。
期刊介绍:
Computers and Electronics in Agriculture provides international coverage of advancements in computer hardware, software, electronic instrumentation, and control systems applied to agricultural challenges. Encompassing agronomy, horticulture, forestry, aquaculture, and animal farming, the journal publishes original papers, reviews, and applications notes. It explores the use of computers and electronics in plant or animal agricultural production, covering topics like agricultural soils, water, pests, controlled environments, and waste. The scope extends to on-farm post-harvest operations and relevant technologies, including artificial intelligence, sensors, machine vision, robotics, networking, and simulation modeling. Its companion journal, Smart Agricultural Technology, continues the focus on smart applications in production agriculture.