A testing procedure including a data collection procedure and a contrastive learning-based approach, for establishing a profile for utility poles surveyed in an embedding space. Unique properties of utility poles are preserved in a low-dimensional feature vector. Similarities between pairs of samples collected at the same or different poles is reflected by the Euclidean distance between the pole embeddings. During data collection—variabilities of excitation signals are manually introduced, e.g. impact strength, impact locations, impact time ambiguity, data collecting location ambiguity on a DFOS/DAS optical sensor fiber/cable. Data so collected provides a learned model learned complete information about a utility pole and is more robust with respect to uncontrollable factors during operation. A model training procedure that effectively extracts a utility pole intrinsic properties (e.g., structure integrity, dimensions, structure variety) and remote extrinsic influence (e.g., excitation strength, weather conditions, road traffic), without knowing the ground truth of these factors. The only identifying label required is an ID of any tested poles, which is readily available. The model is trained adaptively—end-to-end—is advantageously easy-to-implement on modern deep learning frameworks such as PyTorch.