This study introduces a modified deep convolutional neural network model (MDCNN) for recognizing and localizing forest fires in video imagery. The MDCNN model is designed to improve the accuracy of flame detection by integrating a deep CNN with an original feature fusion algorithm. The model employs transfer learning to refine its performance and is trained on a diverse dataset of fire and non-fire scenarios. The proposed model demonstrates a low false alarm rate of 0.563%, a false positive rate of 12.7%, a false negative rate of 5.3%, and a recall rate of 95.4%, achieving an overall accuracy of 95.8%. The experimental results show that the MDCNN model significantly improves the accuracy of flame recognition and exhibits strong generalization ability. The study also addresses the issue of imprecise detection of flame characteristics by integrating a deep CNN with an original feature fusion algorithm. The model's effectiveness is further validated through anti-interference experiments, which demonstrate its ability to accurately recognize static fire scenes as non-fire scenes. Overall, the MDCNN model offers a reliable and accurate solution for forest fire detection and early warning systems.This study introduces a modified deep convolutional neural network model (MDCNN) for recognizing and localizing forest fires in video imagery. The MDCNN model is designed to improve the accuracy of flame detection by integrating a deep CNN with an original feature fusion algorithm. The model employs transfer learning to refine its performance and is trained on a diverse dataset of fire and non-fire scenarios. The proposed model demonstrates a low false alarm rate of 0.563%, a false positive rate of 12.7%, a false negative rate of 5.3%, and a recall rate of 95.4%, achieving an overall accuracy of 95.8%. The experimental results show that the MDCNN model significantly improves the accuracy of flame recognition and exhibits strong generalization ability. The study also addresses the issue of imprecise detection of flame characteristics by integrating a deep CNN with an original feature fusion algorithm. The model's effectiveness is further validated through anti-interference experiments, which demonstrate its ability to accurately recognize static fire scenes as non-fire scenes. Overall, the MDCNN model offers a reliable and accurate solution for forest fire detection and early warning systems.