Aiming at the problems of low detection accuracy resulting from insufficient feature extraction and inaccurate detection box positioning in the Faster R-CNN algorithm, an object detection algorithm based on multiscale feature fusion and anchor adaptation is proposed. First, the high- and low-level features between adjacent levels were fully extracted using the two-way fusion method; then, the multiscale features were balanced so that the integrated features could obtain the same amount of semantic information and detailed information with different resolutions, improving the object recognition ability. Finally, the anchor was generated by adaptively predicting the position and shape of the anchor using the characteristic information of the object in the region proposals network(RPN). The experimental results of the algorithm based on VOC dataset show that compared with the Faster R-CNN algorithm based on ResNet50, the multiscale feature fusion strategy in the proposed algorithm strengthens the detection ability for objects with different scales. The adaptive anchor mechanism can improve the positioning accuracy and avoid missed detection of small objects, and the overall detection results of the proposed algorithm have good performances. The proposed algorithm improves the average detection accuracy by approximately 3.20 percentage points.