Indoor scene recognition based on YOLOv8 improved loss function

With economic growth, indoor image’s application is being used more and more in the community. Because indoor scenes are too complex, and there are few technologies for indoor scene classification, a deep learning network framework YOLOv8-IIR was proposed, which is based on YOLOv8 and improved loss function for indoor scene classification. The original loss function was changed to Focal-EIoU in this article, so the geometric factor differences of the area where the rectangles intersect, their position and respective sizes in bounding box regression (BBR) were measured specifically from it, and the regression is observed in high quality anchor boxes, which is also maked from it. Furthermore, Ghost Convolution Deep and Cross Network(DCN) and Contextual Transformer (CoT) block were added on the basis of YOLOv8. Through the above improvement strategy, experiments were carried out on two data sets of home scene and office scene. Based on YOLOv8, the classification accuracy of YOLOv8_IIR has increased by 7.48%, so the accuracy of indoor scene recognition has been effectively improved.