A structured literature review of computer vision methods for insect identification
Open Access
- 1 July 2026
- journal article
- review article
- Published by Oxford University Press (OUP) in Journal of Insect Science
- Vol. 26 (4)
Abstract
Automated image-based identification of adult insects is increasingly critical to biodiversity monitoring, pest management, and vector surveillance, yet practical deployment remains limited by data scarcity, field variability, and fine-grained taxonomic challenges. We conducted a PRISMA-guided literature review of computer-vision methods for insect classification and identification. A Web of Science Core Collection search (31 August 2024) retrieved 930 records; after deduplication (n = 2) and screening, 230 articles underwent full-text quality assessment using weighted criteria for taxonomy/methods, image capture, computational technique, sample size, and performance evaluation. Of these, 111 high-quality studies met inclusion thresholds. Data were extracted on taxonomic coverage, optical devices and experimental settings, algorithms and pipelines, datasets, and outcome metrics. Deep learning dominated the field; You Only Look Once variants were common for detection and ResNet/EfficientNet/MobileNet for classification; occasional hybrids combined Convolutional Neural Network (CNN) features with traditional classifiers. CNN-based and 1-stage detectors outperformed hand-crafted pipelines; transformers and self-supervised pre-training showed promise with limited labels. Despite strong laboratory performance, generalization to field conditions was hindered by illumination, occlusion, and pose variability. Public datasets were scarce and geographically skewed, limiting reproducibility and equitable benchmarking. Taxonomic coverage concentrated on Lepidoptera, Diptera, Hemiptera, and Coleoptera. We recommend advancing the field through comprehensive reporting beyond overall accuracy, the design of lifecycle-aware and domain-adapted models validated under field conditions, the establishment of diverse benchmarks with standardized imaging protocols, and the development of interpretable architectures suitable for deployment in embedded trapping systems.Keywords
This publication has 85 references indexed in Scilit:
- Detection and recognition of the invasive species, Hylurgus ligniperda, in traps, based on a cascaded convolution neural networkPest Management Science, 2024
- EResNet‐SVM: an overfitting‐relieved deep learning model for recognition of plant diseases and pestsJournal of the Science of Food and Agriculture, 2024
- A YOLO-Based Insect Detection: Potential Use of Small Multirotor Unmanned Aerial Vehicles (UAVs) Monitoring2024
- Low Cost Machine Vision for Insect Classification2024
- Advanced multi-class deep learning convolution neural network approach for insect pest classification using TensorFlowInternational Journal of Environmental Science and Technology, 2022
- Maize Insects Classification Through Endoscopic Video Analysis2019
- Automatic in-trap pest detection using deep learning for pheromone-based Dendroctonus valens monitoringBiosystems Engineering, 2018
- Detection of stored-grain insects using deep learningComputers and Electronics in Agriculture, 2018
- Image-based orchard insect automated identification and classification methodComputers and Electronics in Agriculture, 2012
- Automated insect identification through concatenated histograms of local appearance features: feature vector generation and region detection for deformable objectsMachine Vision and Applications, 2007