Objective: To improve the identification accuracy of fruit categories. Methods: An identification method for fruit categories was established based on an improved Faster regional convolutional neural network （Faster R-CNN） model. Firstly, the regularization method was used to attenuate the weight of some high-dimensional parameters to effectively solve the over-fitting problem that may occur in the training process. Then, two loss functions, a likelihood function and a regularization function, were added to the Faster R-CNN framework to optimize the convolution layer and the pooling layer. Additionally, the least square method was utilized to solve the objective function of fruit recognition. Finally, the accuracy, recall rate, precision and F1 score were used to evaluate the fruit identification effect of the trained fruit identification method. Results: The accuracy, precision and recall rates of the proposed method for fruit identification reached 99.69%, 0.996 8 and 0.994 8, respectively. Compared with the other 10 fruit identification methods, the accuracy, precision and recall rate of the proposed method were at least 0.91%, 1.32% and 0.51% higher. Conclusion: The method can realize the accurate recognition of different categories of fruits.