使用深度學習方法識別咖啡葉病害並估計嚴重程度

Abstract

本研究針對全球咖啡產業因咖啡植物病害日益猖獗而面臨的挑戰，這些病害影響了咖啡產量的質量和數量。本研究引入了一種將計算機視覺技術與深度學習模型相結合的方法，用於檢測和分類咖啡病害並估計病害嚴重程度。研究使用了來自不同來源的1086張圖像數據集，包括阿拉比卡和羅布斯塔咖啡葉片圖像。這些圖像通過處理技術增強後，用於訓練和評估深度學習模型YOLO。YOLO深度學習模型以94.2%的總體mAP50分類病害類型。此外，模型以0.1的置信度閾值量化病害嚴重程度，整體精度為69.6%，從而實現對咖啡植物感染的全面評估。該雙層分類系統使農民和專家能夠通過YOLOv8做出明智的決定，在檢測、分類和估計咖啡葉病害的嚴重程度方面，總體準確率達到78.55%。

This study addresses the challenges faced by the global coffee industry due to the increasing prevalence of coffee plant diseases, which affect the quality and quantity of coffee yield. This research introduces an approach integrating computer vision technology with deep learning models to detect and classify coffee diseases and estimate disease severity. A dataset of 1,086 images from various sources, including Arabica and Robusta coffee leaf images was used. These images, augmented with processing techniques, serve as the foundation for training and evaluating deep learning model, YOLO. The YOLO deep learning model classifies disease types with an overall mAP50 of 94.2%. Additionally, the model quantifies disease severity with an overall Precision of 69.6% with a confidence threshold of 0.1, enabling a comprehensive assessment of the infection in coffee plants. This dual-tier classification system empowers farmers and specialists to make informed decisions in detecting, classifying, and estimating the severity of coffee leaf diseases through YOLOv8, achieving an overall accuracy of 78.55%.