A forest fire recognition method based on a modified deep convolutional neural network (MDCNN) is proposed to enhance the accuracy and efficiency of detecting and locating forest fires in video imagery. The MDCNN model integrates transfer learning and a feature fusion algorithm to improve flame detection accuracy. A diverse dataset of fire and non-fire scenarios is used to train the model, achieving a 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%, with an overall accuracy of 95.8%. The model is designed to handle complex features such as smoke and flames, and it is optimized for real-time monitoring and early warning of forest fires. The MDCNN model is trained using the AlexNet architecture, which is suitable for small-scale datasets and has a relatively small number of parameters. The model is evaluated using a dedicated fire database and demonstrates high accuracy in recognizing and locating forest fires. The study also addresses the challenge of interference from similar fire scenarios by adjusting the coordinates of bounding boxes between consecutive frames. The model's performance is compared with other deep learning models, and it shows superior accuracy and generalization capabilities. The results indicate that the MDCNN model is effective in detecting forest fires with high accuracy and low false alarm rates, making it a reliable solution for forest fire monitoring and early warning systems. The study highlights the importance of using advanced deep learning techniques for forest fire detection and emphasizes the need for further research to refine model parameters and improve recognition performance.A forest fire recognition method based on a modified deep convolutional neural network (MDCNN) is proposed to enhance the accuracy and efficiency of detecting and locating forest fires in video imagery. The MDCNN model integrates transfer learning and a feature fusion algorithm to improve flame detection accuracy. A diverse dataset of fire and non-fire scenarios is used to train the model, achieving a 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%, with an overall accuracy of 95.8%. The model is designed to handle complex features such as smoke and flames, and it is optimized for real-time monitoring and early warning of forest fires. The MDCNN model is trained using the AlexNet architecture, which is suitable for small-scale datasets and has a relatively small number of parameters. The model is evaluated using a dedicated fire database and demonstrates high accuracy in recognizing and locating forest fires. The study also addresses the challenge of interference from similar fire scenarios by adjusting the coordinates of bounding boxes between consecutive frames. The model's performance is compared with other deep learning models, and it shows superior accuracy and generalization capabilities. The results indicate that the MDCNN model is effective in detecting forest fires with high accuracy and low false alarm rates, making it a reliable solution for forest fire monitoring and early warning systems. The study highlights the importance of using advanced deep learning techniques for forest fire detection and emphasizes the need for further research to refine model parameters and improve recognition performance.