113

4,800 千元

評估管理-評估管理

2024/02/23

2024/11/30

李後鋒

化學署危害控制組

蔡秋美

國立中興大學

環境用藥

環境節肢動物監測、影像辨識、昆蟲智慧辨識系統研析

Environmental Arthropod Monitoring, Image Recognition, Intelligent Insect Identification System Analysis

歐陽彥杰

陳良慧

22840361 #563

karenchen@nchu.edu.tw

隨著都市化快速發展，人口大量遷移至都市，充足的食物和適宜的棲息地為昆蟲提供了理想的繁殖和生存環境，近年來氣候變遷也對都市害蟲的種類與分布產生了顯著影響。不同害蟲的防治方法各異，自動化且準確辨識害蟲，成為有效防治都市害蟲的關鍵。透過先進的影像辨識技術，人工智慧（AI）能快速且準確地辨識各種害蟲。本計畫建立針對十四種常見都市害蟲的影像辨識模型，透過影像訓練與測試、並將模型經標籤化處理後，辨識準確度達0.9以上。本計畫同時在台灣西部10個地點，利用捕蟑屋與捕蠅紙進行6個月的每月調查。結果顯示，室內環境中的節肢動物組成主要影響因子為緯度變化和黏紙陷阱誘引方式，而非環境類型的不同。南部的物種數明顯多於北部與中部，昆蟲的種類與數量組成受到緯度相關的氣候條件影響。本計畫亦分析居家昆蟲智慧辨識系統的應用模式，訓練模型在不同設備中的應用具有不同效能與成本表現。在輕量化的邊緣設備上，可以低計算成本達到足夠準確度。未來若將此系統整合至手機APP或網頁端，能為民眾提供實用的害蟲辨識工具。我們規劃建置常見節肢動物辨識、生物學資料與防治建議的居家昆蟲資料庫，期望透過多場景應用優化昆蟲辨識的效能與使用體驗。

環境節肢動物監測、影像辨識、昆蟲智慧辨識系統研析

成果報告1224.pdf

10MB

2025/1/31 上午 12:00:00

With rapid urbanization, large-scale migration to cities has led to a concentration of food sources and suitable habitats, providing ideal conditions for insects to breed and thrive. In recent years, climate change has significantly impacted the types and distribution of urban pests. Effective pest control relies on the ability to accurately and automatically identify pests, as different species require different control methods. Advanced image recognition technology, powered by artificial intelligence (AI), enables the rapid and accurate identification of various pests.This project developed image recognition models for 14 common urban pests. After training and testing with labeled image datasets, the models achieved an identification accuracy of over 0.9. Additionally, a six-month survey was conducted at 10 locations in western Taiwan using sticky traps and flypaper. The results indicated that latitude variation and the type of bait used in sticky traps, rather than environmental type, were the primary factors affecting the composition of arthropods in indoor environments. The species diversity in southern Taiwan was significantly higher than in northern and central regions, influenced by latitude-related climatic conditions. The project also analyzed application models for smart insect identification systems in residential settings. Testing revealed varying performance and cost-effectiveness across different devices. Lightweight edge devices achieved sufficient accuracy at low computational costs. Integrating this system into mobile apps or web platforms could provide practical pest identification tools for the general public. Additionally, plans are underway to create a household insect database, which will include identification features, biological information, and pest control recommendations for common arthropods. Through multi-scenario applications, this initiative aims to enhance the effectiveness and user experience of insect identification systems.

Environmental Arthropod Monitoring, Image Recognition, Intelligent Insect Identification System Analysis