The study applies machine vision technology to the production and operation process of energy enterprises, and constructs a fire detection model based on improved YOLOv4 from the real-time monitoring of fire emergency safety scenarios. Based on the original YOLOv4 algorithm, the model lightens the feature extraction and feature fusion networks, and introduces CA attention mechanism in the bottom layer of the feature extraction network to improve the accuracy of target detection. An intelligent fire alarm system is built on this basis as a response method for emergency security scenarios. Comparison with the basic YOLOv4 algorithm reveals that the improved YOLOv4 algorithm reduces the parameter amount by 45.97%, improves the FPS by 27.75, and improves the mAP value by 14.10%, which achieves a better detection accuracy on the basis of greatly reducing the amount of computation and parameter count, and also achieves a better Loss value and mAP in the comparison with other detection methods. Intelligent Fire Alarm The system integrates intelligent detection, intelligent alarm, intelligent alarm receiving and intelligent alarm dispatching, and can complete the fire alarm process within 6s. In summary, it shows that the method proposed in this paper can be used in real-time monitoring of emergency security scenarios and can provide timely warning at the early stage of security hazards.