Forest fires pose significant risks to ecosystems, biodiversity, and human life, exacerbated by climate change and extreme weather events. Traditional detection methods often face limitations in efficiency, cost, and adaptability to complex environments. This study introduces SE-VGG16, a modified VGG16-based convolutional neural network (CNN) enhanced with Squeeze-and-Excitation Blocks (SEBlocks). The proposed model aims to improve feature representation, enabling accurate detection of small-scale flames and smoke under dense vegetation and varying environmental conditions. We evaluated the model performance using a diverse Kaggle dataset, demonstrating superior accuracy (98%), recall (98.5%), and F1 score (98%), outperforming existing solutions while maintaining computational efficiency. Our approach advances real-time, scalable, and robust forest fire detection, with potential applications in diverse and complex environments, particularly using unmanned aerial vehicles (UAVs).