During the catastrophic failure process, the landslide mass emits low-frequency infrasonic waves, which are characterized by strong penetrating power, low energy attenuation, and long propagation distance, providing a basis for the long-range passive monitoring of the landslide infrasound signal. However, current landslide infrasound monitoring technologies are affected by environmental interference noise and frequently produce false positives. To improve the accuracy of landslide infrasound signal recognition, the monitoring signal needs to be analyzed to determine whether it is a landslide infrasound signal. To this end, this study collected numerous infrasound signals generated in the failure processes of landslide masses of different soil types under different degrees of consolidation through laboratory landslide simulation tests. Furthermore, various types of environmental interference infrasound signals in mountainous areas were gathered by field observations. These signals were divided randomly into training sets and test sets according to a ratio of 3:2. Through the feature analysis of the training set data, the typical features of the landslide infrasound and the environmental interference infrasound in both time and frequency domains were summarized. By constructing the feature vector set and regularization process, as well as using technical means such as the K-nearest neighbor (KNN) classification algorithm, Python, Matlab, and database, an intelligent landslide infrasound signal recognition system was developed. The performance of the recognition system was verified using the test set data. The verification results showed that the system has high recognition accuracy and computational efficiency and can meet the accuracy and real-time requirements of landslide infrasound monitoring. In addition, the recognition results of the system can provide an accurate signal source and reliable information support for landslide infrasound early warning.