In the passive radio frequency identification systems, dynamic frame slotted ALOHA framework has been popularly deployed by the industry driven EPCGlobal C1G2 standard to solve tags collision problem, where tags collision is mainly caused by the mismatched frame length leading to simultaneous responding tags in one same time slot and one reader needs to continuously select the appropriate frame length for effectively identifying tags. Obviously, the throughput improvement comes at the expense of frequent adjustments leading to large computation load and consumption. In order to decrease the frame length adjustment times and catch hold of the satisfactory throughput, this paper proposes a segment-by-segment ALOHA algorithm, where one frame is composed of slot-segments and each slot-segment is composed of s L continuous time slots with three scenarios as collision occupant, empty occupant and singleton occupant. To count these three scenarios in n L slot-segments, the corresponding adjustment operations with exclusive estimator to deal with the unread tags is further introduced. Compared with the state-of-the-art ALOHA-based algorithm in slot-by-slot fashion, the proposed one dramatically decreases the frame length adjustment times and partly increases the identification speed up to 420 tags/s with the throughput around 36% which is very close to the theoretical maximum 36.8%.