Image Classification Task Automation Using Neural Networks

The task of pattern recognition with subsequent classification of the obtained images is considered. The relevance of the topic is determined by the increasing need for automation of pattern recognition processes using advanced technologies such as neural networks, which will significantly improve accuracy and efficiency in solving scientific and production tasks in any sphere of economy. Factors are named, tools and architectures that should be considered when building neural networks are listed, as the speed of code writing and quality of functioning depend on it. It is suggested to use TensorFlow, Keras as deep learning frameworks for implementation. The paper focuses on testing and evaluating the quality of neural network training using various metrics. The paper touches upon the problem of overtraining of neural networks and lists classical ways to prevent overtraining, with emphasis on adding a regularisation layer Dropout. An example of application development using artificial intelligence technologies is given. This kind of applications will automate and greatly facilitate the task of product quality control, reducing image classification time, increasing productivity.