Object Detection from the Images of Hotel Scenario Based on Faster-R-CNN-FFS Model

This paper aims to solve the problem of object detection in images generated by hotel consumers, which have complex backgrounds and various sizes of objects. It proposes an improved Faster R-CNN-FFS model, which improves the Faster R-CNN model from three aspects: first, it uses feature fusion technology based on Resnet-50 to build feature maps, capturing the whole picture semantic and enhancing the detection effect of small objects; second, it adopts Focal loss function instead of cross entropy loss function, increasing the weight ratio of data samples with less quantity, alleviating the problem of imbalanced positive and negative samples of all categories of detecting objects; third, it introduces Soft-NMS mechanism to filter candidate boxes, improving the robustness of object detection model under occlusion scenario. Experimental results show that the recognition performance of the proposed Faster R-CNNFFS model is significantly different from the state-of-the-art models Retinanet, YOLO v3 and Faster R-CNN, and outperforms Retinanet model significantly.