Traditional industrial automatic sorting method is confronted with the problems such as low accuracy of workpiece identification and complex definition of features． Although deep-learning provides a good solution to such problems，there remains the problem of high computational requirements for edge-end equipment． Thus a deep- learning recognition algorithm based on fog and cloud computing was proposed，in which an improved ALEXNET convolution neural network ( CNN) is used for training in the cloud． Then，the trained CNN is downloaded to the fog ( edge) device to recognize workpieces in real time． The results of the experiment carried on 100 different workpieces show that accuracy of the proposed workpiece recognition algorithm is improved from 98% of ALEXNET to 99% ，and the model parameters are reduced by 25% ． The workpiece recognition algorithm can fully utilize the powerful cloud computing capabilities and the real-time nature of edge devices，and provide a novel way for intelli- gent recognition of workpieces．