In this research, we introduce a novel fire and smoke detection system, using the YOLOv7 deep learning model, for industrial use. To reduce high false alarm rates and the difficulty of detecting invisible flames in traditional approaches, the system is trained using a bespoke dataset from Roboflow, which includes more than 11,000 images of fire, smoke, and invisible fire. The data used to train the model was highly augmented. The YOLOv7-based system achieves 90% precision and 84% recall with a mean Average Precision (mAP) of 86% at an Intersection over Union (IoU) threshold of 0.5 with a processing speed of 30 frames per second and less than 2% false alarms. The system is then compared to existing models in terms of its improved accuracy, detection speed, and environmental disturbance resilience. This research contributes to workplace safety by providing a reliable, real-time detection system for both apparent and concealed fires which dramatically improves early warning and reduces risks.