基于YOLOv12的智慧化茶叶病害检测系统的研究与应用

Research and Application of Intelligent Tea Disease Detection System Based on YOLOv12

摘要: 为解决传统茶叶病害检测依赖人工、效率低下且泛化能力弱的问题，本研究构建了一套集高精度检测与智能分析于一体的智慧化系统。技术上，系统以轻量化的YOLOv12模型为核心检测算法，并采用Dash框架开发前端交互平台。在智能分析模块，系统集成了本地部署的DeepSeek-R1-14B大语言模型，通过连接病害数据库(D1)与茶叶知识库(D2)，实现了从数据采集、实时检测到智能诊断与决策的闭环管理。为保证数据源质量，系统采用了IP65/67防护等级的ace2Basler工业相机进行图像采集。实验结果表明，该YOLOv12模型在茶藻斑病、茶褐枯病和茶灰枯病三种病害数据集上的mAP@0.5达到了0.955。同时，模型参数量仅为2.56 M，GFLOPs为6.3，端到端检测速度达到189.30 FPS，实现了高时效性。本方案完成了一个兼顾数据安全与检测效率的智慧化农业系统的开发任务。

Abstract: To address the low efficiency, subjectivity, and poor generalization of traditional manual inspection for tea diseases, this study constructs an integrated intelligent system combining high-precision detection with smart analysis. Technically, the system utilizes the lightweight YOLOv12 model as the core detection algorithm and employs the Dash framework for the front-end interactive platform. For the intelligent analysis module, the system integrates a locally deployed DeepSeek-R1-14B Large Language Model. This LLM connects to a disease database(D1) and a tea knowledge base(D2), enabling a closed-loop management process from data acquisition and real-time detection to intelligent diagnosis and decision-making. To ensure data source quality, an IP65/67-rated ace2 Basler industrial camera is used for image acquisition. Experimental results show that the YOLOv12 model achieved a mAP@0.5 of 0.955 on the dataset featuring three disease types (algal-leaf-spot, brown-blight, grey-blight). Concurrently, the model demonstrates high efficiency, with only 2.56 M parameters, 6.3 GFLOPs, and an end-to-end detection speed of 189.30 FPS. This project (YOLOv12 + DeepSeek + Dash) has accomplished the construction task of an intelligent agricultural system that takes into account both data security and detection efficiency.