The invention discloses a counter-reconnaissance escape attack method for a deep pedestrian re-identification system. A matching difference maximization-based optimization method is proposed, and cross-camera variable and position-expandable noise patterns are generated in combination with multi-position sampling, so that the same noise patterns cannot be matched with each other under shooting ofdifferent cameras at any position of a pedestrian re-identification system monitoring area. Besides, physical environment factors are fused into the noise pattern generation process, information lossof noise in the printing and shooting process is reduced, and the robustness of the method is improved. The noise pattern generated by the method can enable a pedestrian re-identification system not to correctly search and locate an attacker, thereby realizing invisibility under a security monitoring system.