Percorrer por autor "Santos, Pedro Albuquerque"
A mostrar 1 - 2 de 2
Resultados por página
Opções de ordenação
- AI-powered solution for plant disease detection in viticulturePublication . Madeira, Miguel; Porfírio, Rui; Santos, Pedro Albuquerque; Madeira, Rui NevesIn an era dominated by the intersection of advanced technology and traditional industries, the domain of agriculture is on the verge of a revolutionary transformation. This article introduces a solution for vineyard producers, harnessing satellite imagery, weather data, and deep learning (DL) to identify vineyard diseases robustly. This solution, designed for proactive plant health management, stands as a transformative tool towards digital viticulture. Such tools transition from luxuries to essentials as vineyards confront evolving challenges like climate change and new pathogens. Our research builds on the hypothesis that customising deep learning architectures for specific tasks is crucial in enhancing their effectiveness. We contribute by introducing a tailored convolutional neural network (CNN) architecture, developed specifically for the classification of plant diseases using vineyard imagery. The experimental results demonstrate that our custom CNN architecture exhibits performance on par with established state-of-the-art models like ResNet50 and MobileNetV2, underscoring the value of specialized solutions in addressing the unique challenges of viticulture. This paper introduces an overview of the solution’s architecture, presents the implementation of DL modules with their corresponding results, and describes use case scenarios.
- Exploring explainable AI techniques for plant disease classification in digital agriculturePublication . Porfírio, Rui Pedro; Madeira, Rui Neves; Santos, Pedro AlbuquerqueIntegrating smart farming technologies in agriculture is crucial to address the pressing challenges of food security, economic stability, and environmental sustainability. Solutions based on artificial intelligence (AI) for plant disease detection play a critical role in optimizing crop health and yield. However, the complexity and opacity of these AI models can hinder their acceptance and practical application by end-users. To address this issue, our ongoing research explores applying explainable AI (XAI) techniques to enhance the explainability of vision-based plant disease classification models. Our experiments assess the transparency of vision-based models by applying XAI techniques, such as LIME, Grad-CAM, and occlusion-based attribution, to visualize the reasoning behind model predictions. Additionally, we analyze inherently transparent machine learning models, such as k-Nearest Neighbors, through custom visualization graphics and examine how explainability varies with model accuracy. These findings highlight the role of XAI techniques in enhancing the transparency of predictions for crucial vision tasks in agriculture such as plant disease classification. Building on these results, we explore the potential integration of X AI techniques into third-party applications or prototypes, emphasizing how tailored visual and textual explanations can enhance the transparency and interpretability of plant disease classification models.