Wildfires pose a significant threat to ecosystems, biodiversity, and human communities. Early detection of high-risk areas is crucial for implementing preventive measures and mitigating damage. This research proposes an approach based on image processing techniques and convolutional neural networks (CNN) to identify areas prone to wildfires. Multiple neural network architectures, such as YOLOv8, YOLOv11, and U-Net, were evaluated using a dataset of satellite images. The models were trained and validated using precision, recall, and F1-score metrics. The results indicate that integrating neural networks and image processing is effective for the early detection of high-risk areas. The YOLOv11m-Seg model demonstrated an accuracy of 93.4% and a mean average precision (mAP) of 98%, highlighting its strong capability to accurately identify and classify wildfire risk zones. This study contributes significantly to the development of computational tools aimed at wildfire prevention, providing a scalable and precise approach suitable for implementation in real-time monitoring systems. By leveraging advanced algorithms and data-driven methodologies, the proposed solution enhances the capability to detect, assess, and respond to potential wildfire threats efficiently. In addition, the study involves the generation of a labeled dataset that classifies forest areas according to their level of fire risk. This dataset not only supports the validation and training of predictive models but also ensures the practical applicability of the system in the field, enabling authorities and stakeholders to make informed decisions regarding forest management and emergency response.