The work aims to propose an object detection method based on attention and feature fusion (AT-FFRCNN) aiming at the problems of different size, mutual occlusion and different working forms of construction machinery objects in the construction environment.ResNet50 and feature path aggregation network PFPN were used in the backbone network to fuse feature information of different scales, and an attention mechanism was introduced into the region proposal network (RPN) and fully connected layer to improve the ability of target recognition, and generalized intersection over union (GIoU) was used in the loss function to improve the accuracy of the object box.Experiments indicated that the detection accuracy of the proposed method was greatly improved compared with other methods, and the average detection accuracy (mAP) reached more than 90%. The proposed method can complete the detection task of the construction machinery better.