In forensic DNA analysis, mixed STR profiles with unknown number of contributors remain one of the most challenging research problems. The complexity and difficulty of analyzing mixed STR profiles increase exponentially with the number of contributors, and traditional methods, which typically rely on fixed model structures, struggle to adapt to this complexity. Based on the Deformable Convolutional Network (DCN), this paper proposes a kind of variable structure network for STR Profiles Analysis—VSNSTR. This method utilizes convolutional layers to extract STR profile features, applies an additional convolutional layer to the convolutional features to obtain deformation offsets for the deformable convolutions, enabling the feature extraction to automatically adjust its scale or receptive field based on complexity; introduces a dynamic routing mechanism to perform separate complexity analysis on the output features of each layer, enabling the neural network to dynamically adjust its network branches and computational depth based on complexity. This avoids overfitting of high-complexity models on low-order mixed samples and the insufficient representational capacity of low-complexity models on high-order samples. Therefore, this method can adaptively learn peak-shaped features under different mixture structures without the need to predefine the number of contributors, thereby improving the accuracy of contributor structure discrimination. Experimental results demonstrate that by constructing a deep learning framework capable of dynamically adjusting the model structure, this study achieves higher recognition accuracy and stability under conditions of unknown contributor numbers, enabling the automatic analysis of STR profiles under varying contributor conditions and providing a new technical solution for the automatic analysis of complex DNA mixture profiles.

Authors

• Haoyu Cui (School of Cyberspace Science and Technology, Beijing Institute of Technology, China)
• Liang Zeng (School of Cyberspace Science and Technology, Beijing Institute of Technology, China)
• Zhen Xu (Institute of Forensic Science, Ministry of Public Security, China)
• Fan Yang (Institute of Forensic Science, Ministry of Public Security, China)