Fingerprint indoor localization technology based on Wi-Fi has been widely used due to its good accuracy and easy implementation. Most of them use Received Signal Strength (RSS) as fingerprint because RSS is easy to obtain, but due to the poor stability of RSS, CSI has gradually become the mainstream of fingerprint due to its excellent stability. Traditional fingerprint location uses KNN or Bayes for fingerprint matching. On the one hand, it has high computational complexity, on the other hand, there is a large error in positioning accuracy. With the development of deep learning, combining fingerprint location with a neural network, using the excellent feature extraction ability of the neural network, we can extract higher dimensional signal features of Wi-Fi signal, and further improve the positioning accuracy. In order to improve the performance of Wi-Fi fingerprint indoor localization technology, a channel state information (CSI) fingerprint indoor localization method based on convolutional neural networks (CNN) is proposed. This method uses the amplitude difference and phase difference information of the reference point of the positioning environment as the offline training data, and saves the CNN network model as the fingerprint; In the online phase, according to different experimental scenarios, the amplitude difference information and phase difference information of the test data are weighted, and an improved probability-based fingerprint matching algorithm is introduced to estimates the coordinates of the test points by CNN network model. In addition, in order to enhance the universality of the algorithm, a two-node localization scheme is proposed to improve the positioning accuracy for complex indoor scenes. Experiments are carried out in two different positioning scenarios: the corridor and the laboratory. The average positioning error of the information joint localization algorithm is 28.4 cm and 65.3 cm respectively, which verifies the effectiveness of the proposed algorithm.