In order to study the detection and recognition of rotating machinery, the attributes of each task and their relations, a joint detection and recognition algorithm for coarse-grained attributes was proposed. The algorithm of vehicle brand recognition was studied especially for fine-grained attribute recognition. First, the colour and type of the rotating machine were fused into the detection algorithm. The multi-task learning framework was used to model the attribute recognition task and positioning task of rotating machinery. At the same time of detection, attribute recognition was completed. Then, the deep learning method (DL) and convolution neural networks (CNN) were introduced, which was widely used characteristics of deep learning approach. Combined with its structural characteristics and generalization capabilities, the CNN structure was analyzed. Finally, experimental tests were conducted. The universal validity and environmental adaptability of the proposed detection algorithm were verified. The results showed that based on the rotating mechanical dataset, the proposed rotating machinery recognition algorithm not only accurately identified the known class samples in the test set, but also identified the samples of unknown categories. Therefore, the proposed rotating machinery detection and identification framework solves the problems in the current solution. The effect of rotating machine detection and recognition is enhanced. The overhead of solution computing resources is reduced. This has practical application value.