Abstract: To address the severe challenges faced by civilian explosive production enterprises in the safety management of personnel and vehicles，and in response to the shortcomings of traditional access con⁃ trol systems such as lax identity verification，data silos，and lack of real-time functionality，a high-securi⁃ ty and high-efficiency barrier-free intelligent access control system has been developed. In terms of re⁃ search methods，the system is centered on the improved FaceNet deep learning model. By constructing a dedicated dataset for the civilian explosive scenario and adopting transfer learning and network lightweight⁃ ing technologies，the recognition accuracy and speed of the algorithm under complex conditions (such as wearing safety helmets/safety masks，dust，and changes in light) have been improved. At the same time， the system integrates industrial-grade high-definition cameras，infrared grating detection，sound and light alarms，and LED information screens，and designs a software platform with six core functions including non-intrusive passage，real-time personnel counting，over-capacity alarm，and historical traceability. The research results show that after continuous field tests in multiple scenarios，the system has a face rec⁃ ognition success rate of up to 99.1%，a single-person response time of only 0.35 s，and a concurrent pro⁃ cessing capacity of 300 people per minute. All performance indicators are significantly superior to those of traditional access control systems. The conclusion is that this barrier-free access control system successful⁃ly achieves precise，efficient，and intelligent management of personnel and vehicles in the core areas of ci⁃ vilian explosive production，effectively eliminating security risks such as identity fraud and over-capacity operations，and providing reliable technical support and practical examples for the digital transformation of safety in the civilian explosive industry.

Key words: safety management and control, access control system, image recognition, face recogni? tion