To solve the problem of low efficiency and error prone of manual feeding of the crankshaft bearing caps (CBCs), the visual location and detection method of CBC feeding robot based on attention mechanism is studied to realize automatic feeding. Aiming at the unapparent image features, the attention mechanism is introduced into the feature extraction network of Faster R-CNN to map the weights of the CBC image at different positions to the feature channel, so that the deep learning model can pay more attention to the edge and center semantic information of the CBC. To further improve the location accuracy, this paper also improves the candidate box generation method and loss function. Experiment results show that compared with those of traditional machine learning methods and classic deep learning target detection models, the detection speed of this method reaches 0. 419s, the location accuracies are the best (IOU and GIOU are 0. 941 3 and 0. 940 9, respectively). In addition, the proposed method possesses good robustness. On site test shows that the success rate for the method guide the feeding robot to grasp and place the CBCs reaches 95. 14% , which improves the efficiency of the engine assembly line.