Structural health monitoring is a crucial approach for ensuring the safety and integrity of composite material structures in aircraft. Distributed fiber optic sensors based on backscattered Rayleigh scattering provide data support for composite material damage monitoring by measuring high-density strain distributions. However, the mapping relationship between structural strain distribution characteristics and damage is complex, making it challenging to accurately determine the quantitative information of damage based solely on strain distribution. Additionally, the large volume of data from distributed fiber optic sensors makes manual analysis of strain data time-consuming and less accurate. To address this challenge, an intelligent damage identification method based on distributed fiber optic sensing data and the U-Net neural network is proposed. It aims to automate the precise identification of common delamination damage in composite materials. Initially, training and validation sets for the U-Net neural network are constructed through finite element simulations. Subsequently, cantilever loading tests of composite material plates with delamination damage are conducted, and structural strain distribution data are collected as a test set using distributed fiber optic sensors. The damage identification results demonstrate that the U-Net neural network can accurately quantify the position, size, and shape of delamination damage.