Method of identification of individuals based on an in-depth neural face algorithm adaptive Face Recognition in the wild under heavy noise

The robot is based on the AFRW-X technique, which is based on the idea of integrating adaptive diffusion filtration, implemented on two levels of abstraction, with a pseudo-3D mechanism projection to ensure effective multi-angle vector presentation. The key components of the proposed architecture are: the cascade mechanism of image renewal, the Peron–Malik anisotropic diffusion mechanism, and the composite waste function. The cascade mechanism is characterized by automatic selection of strategy depending on the type of degradation (ROF-TV for moderate noise, PnP-ADMM for mixed degradations and DPS for extreme minds). The anisotropic diffusion mechanism is based on deep character maps from virtual angles and vicoristic adaptive value based on three-component metrics (visibility, IoU, character brightness). The composite spending function, in turn, integrates the class balancing mechanism. Experimental validation on four standard datasets confirmed the superiority of the proposed method in the minds of multiple degradation of input images. Thus, with a critical level of additive Gaussian noise, the AFRW–X (modified) method will ensure excellent recognition accuracy at an extremely low level of variable processing, which exceeds the indicators of the basic ArcFace and the original AFRW versions. The key indicator is an insignificant drop in accuracy compared to strong noise, which satisfies the established possibility of an acceptable reduction and confirms the high robustness of the technique. In the context of complex degradation, which consumes all negative factors, a solution has been developed that achieves an accuracy of 88.5 %, recognition with a low level of agricultural products and high performance The integral metric is 78.5 %, which significantly exceeds the results of other methods. Scientific novelty lies in the synthesis of adaptive cascade filtration, an advanced pseudo-3D projection module and a composite cost function for one-hour compensation of multiple degradation factors. This approach will ensure comprehensive compensation for the multiple degradation of input biometric data, in addition to other methods that look at other negative factors isolated.