Arquitectura avanzada de aprendizaje profundo para la detección de plagas camufladas en agricultura de precisión

Abstract

This thesis introduces AINet, a novel deep learning architecture designed for detecting camouflaged objects in complex and diverse environments. AINet leverages the strengths of Mamba, an efficient sequential state model for capturing long-range dependencies, and the Convolutional Block Attention Module (CBAM) for feature refinement through attention mechanisms. Detecting camouflaged objects is a significant challenge across a wide range of real-world applications, including surveillance, security, medical imaging, and autonomous systems, where objects of interest may blend into their backgrounds and evade conventional detection methods. To demonstrate its effectiveness, AINet is evaluated on multiple datasets, including standard camouflaged object detection benchmarks such as CAMO, COD10K, and NC4K, as well as pest detection datasets. Experimental results show that AINet outperforms existing state-of-the-art models. The code is publicly available on GitHub for reproducibility: https://github.com/hvelesaca/AINet.