This article explores the performance of three convolutional neural network (CNN) models—MobileNet, EfficientNetB0, and InceptionV3—on different image classification tasks with limited computational resources. The study investigates how quickly these models can be trained using varying training bases, including simple color recognition, object identification, and complex image classification. The models were tested on three different training databases, with the first focusing on color recognition, the second on identifying red and blue cuboids, and the third on distinguishing electrical elements from other objects. The results show that MobileNet achieved high accuracy (up to 98%) with a relatively short training time, while InceptionV3 provided the highest accuracy (99.7%) but required significantly longer training. EfficientNetB0 showed comparable performance to MobileNet but with slightly longer training times. The study also demonstrates the feasibility of deploying these models on programmable logic controllers (PLCs) using Siemens' NPU module and Intel RealSense camera, achieving a 90% success rate in image classification within 180 seconds. The research highlights the effectiveness of these models in industrial applications, particularly in scenarios with limited computational resources. The findings suggest that MobileNet is a suitable choice for real-time image recognition tasks due to its balance of accuracy and efficiency, while InceptionV3 offers higher accuracy at the cost of increased training time. The study underscores the potential of deep learning models in industrial automation and the importance of selecting appropriate models based on specific application requirements and hardware constraints.This article explores the performance of three convolutional neural network (CNN) models—MobileNet, EfficientNetB0, and InceptionV3—on different image classification tasks with limited computational resources. The study investigates how quickly these models can be trained using varying training bases, including simple color recognition, object identification, and complex image classification. The models were tested on three different training databases, with the first focusing on color recognition, the second on identifying red and blue cuboids, and the third on distinguishing electrical elements from other objects. The results show that MobileNet achieved high accuracy (up to 98%) with a relatively short training time, while InceptionV3 provided the highest accuracy (99.7%) but required significantly longer training. EfficientNetB0 showed comparable performance to MobileNet but with slightly longer training times. The study also demonstrates the feasibility of deploying these models on programmable logic controllers (PLCs) using Siemens' NPU module and Intel RealSense camera, achieving a 90% success rate in image classification within 180 seconds. The research highlights the effectiveness of these models in industrial applications, particularly in scenarios with limited computational resources. The findings suggest that MobileNet is a suitable choice for real-time image recognition tasks due to its balance of accuracy and efficiency, while InceptionV3 offers higher accuracy at the cost of increased training time. The study underscores the potential of deep learning models in industrial automation and the importance of selecting appropriate models based on specific application requirements and hardware constraints.