EVALUATION OF THE EFFECTIVENESS OF NEURAL NETWORK ALGORITHMS WHEN WORKING WITH DIGITAL IMAGES

Ryzhkova Elena Vladimirovna, Titov Dmitry Nikolaevich

Modern trends in production modernization are impossible without the use of new technological solutions that allow increasing the speed of defect recognition and, as a result, accelerating decision-making. The study is devoted to a comparative analysis of the effectiveness of neural network algorithms in solving the problem of recognizing defects on a product [1, p. 83]. This paper presents an approach to image classification using a pre-trained model. The main training method is Transfer Learning, which allows you to use a pre-trained model (VGG16) and retrain it on your own data [2, p. 95]. The use of cross-validation increased the ability of the model to correctly classify images with defects, increased confidence in the presence of defects in the obtained images. The experiment includes the use of various algorithms that allow you to analyze their work by analyzing graphs and success rates. The model effectively solves the problem of detecting defects in the form of cracks in images, achieving high accuracy in defect classification.

Keywords:neural network algorithm, trained model, convolutional neural networks, machine vision, defect, quality improvement