In light of the rising risk of fires in recent years, the necessity for better fire detection systems has become increasingly evident. Conventional fire detection systems are often utilized to detect fires, however, they have limitations. Recently the field of computer vision has been revolutionized by the Deep learning algorithms by their exceptional ability to identify patterns and extract information. Utilizing deep learning for vision-based quick fire detection can help overcome sensor constraints, reduce false alarms, and increase coverage efficiency. This paper begins by addressing the significance of fire detection and then delving into a wide variety of methods, including both conventional and cutting-edge deep learning approaches. CNNs are often utilized in the detection of fire images in which models such as YOLO and Faster R-CNN provide real-time detection capabilities. Vision Transformer (ViT) has demonstrated superior performance in image classification relative to Convolutional Neural Networks. Benchmark datasets like the Forest Fire Dataset, FLAME dataset, BowFire, FIRE Dataset, and Foggia Dataset are extensively used for training and evaluating fire detection algorithms. The survey aims to provide researchers with significant insights into state-of-the-art, recent deep learning techniques in this rapidly evolving field by integrating findings from multiple sources.