Aiming at the variability in the number and positional configuration of sensors in plantar pressure acquisition devices within the field of identity recognition, as well as the issue of increased time cost arising from the conventional reliance on complete segmentation of gait cycle data for extracting plantar pressure features, this article proposes an identity recognition method based on threshold-free recurrence plots of plantar pressure signals and the LeNet neural network. Plantar pressure data are collected during natural, unloaded walking from 28 healthy adult participants without foot or lower limb pathologies. Data acquisition occurs on a standardized concrete surface using a custom-designed plantar pressure measurement system. The raw plantar pressure data are preprocessed using a data reconstruction algorithm to directly convert them into threshold-free recurrence plots. These generated images are then used as input to the LeNet network for feature extraction and identity recognition. Recognition performance across various single-region and multi-region configurations is systematically analyzed and compared. Experimental results show that the optimal configuration—combining the medial heel, lateral heel, second metatarsal head, and hallux regions—achieved superior identity recognition performance with minimal sensor deployment and high accuracy. Specifically, accuracy, precision, recall and F1-score attained 99.25%, 99.22%, 99.39%, and 99.26%, respectively. Recognition performance tends to be influenced by gait phase and plantar pressure magnitude. However, this effect progressively diminishes with increasing number of integrated regions. Furthermore, the proposed method maintains high recognition accuracy without requiring rigorous gait segmentation. It provides new ideas and technical support for the application of identity recognition technology in the field of biometric identification, and has potential application value in public safety and other fields.