Timely monitoring of sow oestrus is very important in sow breeding. Recently, recognition methods of sow oestrus are low sensitivity, wasting time and usually affected by environment. To resolve these problems, based on ear erect behavior of large white pigs during estrus, a method of large white sow’s oestrus behavior recognition based on convolutional neural network (CNN) was proposed. A model based on AlexNet convolutional neural network, named AlexNet_Sow was firstly developed. Then, AlexNet_Sow model was simplified to get a new model named AlexNet_Sow_Simplified, which contained two convolution modules and two fully connected modules. The activation function of AlexNet_Sow_Simplified was rectified linear units (ReLU), adaptive moment estimation (Adam) was used to optimize gradient descent, and softmax was used to be the classifier of our model. Ear images of oestrus and non-oestrus large white sows were collected and divided into training data (80%) and testing data (20%). The model was trained by using data augmentation method, the accuracy of testing data was 99%. In addition, it was found that when sows’ ears were erect for 76s during teasing, it could be judged as the symbol of oestrus. In order to verify this method, LabVIEW Python nodes were used to intergrate the AlexNet_Sow_Simplified model and set a time threshold of 76s and verified a set of new photos. The result showed that the precision rate, recall rate and accuracy rate of this method to recognize sow oestrus were 100%, 83.33%, and 93.33%, respectively. The average detecting time of a single image was 26.28ms. It proved that this method could achieve noncontact, automatic, and fast detecting of oestrus in large white sows with high accuracy, which could greatly help to reduce sows’stress and the labor cost.