Internet of Things‐based smart insect monitoring system using a deep neural network. Issue 6 (6th September 2022)

• Record Type: Journal Article
• Authors: Wang, JiangTao; Bu, Yufei
• Is Part Of: IET networks. Volume 11:Issue 6(2022)
• Page Start: 245
• Page End: 256
• Publication Date: 2022-09-06
• Subjects: computer vision -- deep learning -- entomology -- image‐based -- sensor‐based; Computer network architectures -- Periodicals; Computer network protocols -- Periodicals; Information networks -- Periodicals; Telecommunication systems -- Periodicals; 004.605
• DOI: 10.1049/ntw2.12046
• Languages: English
• ISSNs: 2047-4954

Abstract

Abstract: According to recent studies, insects make up the vast majority of all animal species on the planet, and their numbers are rapidly dwindling. This occurrence has been recorded by a variety of insect taxa and geographic locales, but there is insufficient data to identify its true extent. Most monitoring methods are ineffective and time‐consuming, making insect population analysis difficult. Computer vision and deep learning advancements, on the other hand, may soon give new answers to this global challenge. Using cameras and other sensors, entomologists can gather data continuously and non‐invasively at any time of day or season. To capture specimens' physical appearance, automated imaging can be utilised in the laboratory. This data can be used to build a deep learning model to predict the quantity, biomass, and variety of insects. Additionally, deep learning models can measure the variability of phenotypic features, behaviour, and interactions. We may thank current advances in deep learning and computer vision for the immediate demand for more cost‐effective insect and invertebrate monitoring systems. Insect monitoring using sensors is demonstrated. Here, we show how deep‐learning algorithms might extract ecological information from massive data sets, and we explore the challenges ahead. There are four areas that will help us achieve our goal: There is a need for verification of taxonomic identification using pictures; sufficient training data; public reference databases; and ways for integrating molecular and deep learning technology.