Titulo Estágio
Automated Insect Classification Using Computer Vision for Real-World
Áreas de especialidade
Sistemas Inteligentes
Local do Estágio
CISUC
Enquadramento
Accurate classification of small, similar-looking objects in noisy and imperfect images is a fundamental challenge in computer vision. This internship focuses on developing a machine learning–based system for fine-grained classification of small objects, more specifically, insect images captured in field conditions. These objects are difficult to distinguish due to their size, motion blur, low contrast, and variability in lighting or background noise. Such challenges are common in many real-world applications of AI and demand innovative solutions in model design, data preprocessing, and robustness.
As a practical use case, the project focuses on the classification of tiny insect vectors (e.g., biting midges), which are relevant to epidemiological studies and agricultural monitoring.
Objetivo
The goal is to develop and evaluate a deep learning framework capable of identifying and classifying tiny insect species from challenging image data. This involves handling low-resolution input, minimizing the impact of visual noise, and optimizing for high classification accuracy across multiple classes with subtle inter-class differences, using public datasets as https://universe.roboflow.com/p-bxylb/mosquito-fly-culicoides-lab/browse?queryText=&pageSize=50&startingIndex=0&browseQuery=true.
This internship will provide hands-on experience with:
Fine-grained image classification
Transfer learning and model finetuning
Data augmentation for low-quality image scenarios
Evaluation and optimization of model robustness
To achieve the above-defined goal, the following objectives will be pursued:
Conduct a review of state-of-the-art methods in small object detection and classification
Analyze and compare existing ML architectures for fine-grained image recognition
Design, implement, and evaluate a custom framework for insect identification
Explore preprocessing, augmentation, and domain adaptation techniques to improve model performance in noisy environments
Plano de Trabalhos - Semestre 1
Literature review on small object detection and image classification
Analyze available datasets and challenges in insect imagery
Define the ML pipeline and select baseline models (e.g., ResNet, EfficientNet, Vision Transformers)
Begin implementation and preliminary testing
Write an intermediate progress report
Plano de Trabalhos - Semestre 2
Finalize implementation and conduct full evaluation
Experiment with image enhancement, domain adaptation, and model tuning
Analyze results and compare with baseline approaches
Write the final internship report
Condições
This work will take place in the context of a research project in CISUC. There is the possibility of a 6-month scholarship of 798 euros per month.
Observações
This works results from a collaboration with INIAV that will provide the datasets that support this work.
Orientador
Catarina Silva, Bernardete Ribeiro, Joana Costa
catarina@dei.uc.pt 📩