The invention relates to a deep convolutional neural network-based stockyard smoke monitoring and online model updating method. The method comprises the following steps of a, converting a smoke video into an image sequence by using a tool, and conducting the smoke labeling operation; b, by using the deep convolutional neural network, obtaining the more abstract high-level features of the smoke, iteratively training and optimizing model parameters, evaluating an iterative model according to a loss function and selecting an optimal model; c, according to the false-information and misinformation condition, updating the model in the online manner. According to the technical scheme of the invention, the effective feature extraction for smoke images is conducted based on the deep convolutional neural network. Based on the method, the smoke monitoring and the model updating are conducted in real time in the video monitoring condition. Therefore, the detection accuracy is further improved. The method is simple in operation, fast, effective, and high in robustness.