A Novel Detection of Hazardous Fires from CCTV Images using Novel YOLOv7 in Comparison with Faster R-CNN

Main objective of this research is to detect hazardous fire. This works utilizes Novel YOLOv7, and compares its precision and performance with Faster R-CNN. Two groups of 34 samples each totaling 68 samples were collected for this analysis. Faster Region-Based Convolutional Neural Networks (Faster R-CNN) are used in Group 2 while Novel YOLOv7 is used in Group 1. The dataset was initially imported with the assistance of the Roboflow tool, and then, after being trained using the Novel YOLOv7 algorithm with the support of Google Colab, it was exported. The process flow for this investigation includes all of these individual components. The sample size is determined by using the findings of an online statistical tool that has a pretest power of 80% and an alpha value of 0.05. This allows for the sample size to be estimated based on the findings of the earlier research that was carried out. From the results obtained, the Novel YOLOv7 algorithm had a mean average precision (mAP) of 0.4506, and Faster R-CNN obtained an mAP of 0.3062. There is a statistically significant difference in precision between both the algorithms with significance values of 0.000 (p<0.05). For the given dataset, Novel YOLOv7 detects hazardous fires more precisely than Faster R-CNN.