This paper explores the application of three convolutional neural network (CNN) models—MobileNet, EfficientNetB0, and InceptionV3—on limited computer resources. The authors investigate the models' performance across different training bases, including simple color recognition, orthogonal element recognition, and more complex image classification. The study aims to demonstrate how the number of images and epoch types affect the models' capabilities. The models were trained using TensorFlow and Keras libraries on a virtual machine with limited hardware parameters. The results show that the MobileNet model achieved a 90% success rate with a learning time of 180 seconds. The paper also discusses the advantages and disadvantages of each model and highlights the potential for industrial implementation on programmable logical controllers (PLCs). The research is the first step towards realizing image recognition using an S7-1500 family controller and an Intel RealSense camera. The study concludes that the selected models are suitable for low-computer-parameter environments and have significant potential in the industrial sector.This paper explores the application of three convolutional neural network (CNN) models—MobileNet, EfficientNetB0, and InceptionV3—on limited computer resources. The authors investigate the models' performance across different training bases, including simple color recognition, orthogonal element recognition, and more complex image classification. The study aims to demonstrate how the number of images and epoch types affect the models' capabilities. The models were trained using TensorFlow and Keras libraries on a virtual machine with limited hardware parameters. The results show that the MobileNet model achieved a 90% success rate with a learning time of 180 seconds. The paper also discusses the advantages and disadvantages of each model and highlights the potential for industrial implementation on programmable logical controllers (PLCs). The research is the first step towards realizing image recognition using an S7-1500 family controller and an Intel RealSense camera. The study concludes that the selected models are suitable for low-computer-parameter environments and have significant potential in the industrial sector.