The invention provides a real-time video field fire smoke detection method based on a convolutional neural network. A smoke image data set is collected through an experimental simulation mode, and a training set, a test set and a verification set are created; the training set, the test set and the verification set are subjected to automatic annotation, and in combination of manual adjustment, thetraining set, the test set and the verification set with a real label are obtained respectively; the training set and the verification set with the real label are subjected to image rotation processing, color channel color addition and subtraction processing and scaling processing to obtain the processed training set and the processed verification set with the real label; the parameters of the convolutional neural network are initialized, and according to the training set with the real label after scaling processing, a well-built convolutional neural network model is trained; a to-be-detectedfield monitoring image is acquired in real time, and through the trained convolutional neural network model, a smoke target detection frame is predicted and optimized; and inter-frame confidence enhancement and relocation are carried out on the target detection result given by the trained convolutional neural network model.