Abstract: The presence of noise and other isotope characteristics in the background environment can significantly impact the accuracy of thickness measurements when using the γ-absorption method in a noisy or complex radioactive environment. Therefore, it is essential to conduct research on a gamma absorption method with strong anti-noise capabilities and good stability. Aiming at the problems of energy spectrum background elimination and peak area calculation in gamma thickness measurement, a gamma thickness measurement method was established based on U-Net deep learning network to improve the measurement efficiency, stability and anti-interference ability of gamma thickness measurement method. Firstly, a gamma absorption thickness measuring device was built based on ²⁴¹Am radioactive source and cadmium zinc telluride detector to measure samples of different thickness and obtain sufficient energy spectrum data. The corresponding baseline elimination spectra were obtained by traditional methods and manual methods, and the training dataset was established. The gamma spectrum baseline elimination network was built based on U-Net, and the network was trained using the established dataset. Using the PPS40 thickness measurement results as an example, Gaussian noise and a Gaussian peak were added to the original energy spectrum data to simulate various levels of noise and radioactive environment. The stability of the thickness measurement method under various noise and radioactive environment was studied, and it was compared with the conventional method. The findings reveal that the established gamma thickness measurement method yields a maximum peak area change rate of only 0.049% under the same noise environment, outperforming the conventional approach’s 0.85%. In the presence of interference from other Gaussian peaks, the proposed gamma thickness measuring method can effectively subtract the Gaussian peaks unrelated to thickness measurement, with a maximum peak area change rate of only 7.58%, outperforming the traditional method which has a peak area change rate of 33.29%. This demonstrates a superior anti-interference capacity compared to the conventional method. As a result, the gamma thickness measurement method based on deep learning offers a novel approach to the data processing of gamma ray thickness measurement. It enhances the gamma thickness measurement method’s anti-noise and anti-interference capability, enabling its application in complex radiation environments and expanding its potential application fields.