Artificial Intelligence in Industry 4.0 and 5G Technology

7Adam Adaptive Optimization Method for Neural Network Models Regression in Image Recognition Tasks

Denis Y. Nartsev¹, Alexander N. Gneushev^1,2, and Ivan A. Matveev²

¹Moscow Institute of Physics and Technology, Department of Control and Applied Mathematics, Institutskiy per., 9, Dolgoprudny, Moscow Region, 141701, Russia

²Federal Research Center “Computer Science and Control” of Russian Academy of Sciences, Dorodnicyn Computing Center, Vavilov str., 40, Moscow, 119333, Russia

7.1 Introduction

Currently, as the most effective approach for solving complex and poorly formalized problems, building models in the field of intelligent systems is the usage of neural network models and machine learning. Such models are applied to the tasks of automatic localization, segmentation, and recognition of objects, identifying a person by the images of a face, iris, hand, etc. In intelligent image analysis systems, image preprocessing is important for the stable extraction of object features, as well as for filtering those images where reliable object recognition cannot be achieved with the required reliability.

In particular, an important step for the task of identifying a person from a face image is preliminary image alignment. To do this, one needs to assess the position of the face in the image. Approaches to solving this problem can be divided into three types: cascade regression of the form [1], statistical models of the form [2–6], and neural network regression models [7,8], evaluating ...