Project start: 1.10.2025.

The current standard in image classification relies on deep convolutional neural networks (CNNs), which require high computational power and GPU acceleration. However, GPUs are expensive, energy-intensive, and impractical for energy-constrained systems such as IoT devices, industrial sensors, and medical diagnostics. This project aims to develop an alternative image classification method based on Genetic Programming (GP), providing high classification accuracy while significantly reducing computational energy consumption.

The first step involves designing a mathematical model for feature extraction, enabling clear class separation without convolutional layers. This step will explore transformations such as matrix decomposition (SVD, PCA), optimization algorithms, and evolutionary techniques to automatically identify relevant features. Following this, classical Genetic Programming (GP) will be applied to construct symbolic models serving as classifiers. The proposed system is designed to operate on CPU architectures, ensuring applicability in energy-efficient systems. Furthermore, the method will be optimized for embedded systems and microcontrollers, opening possibilities for real-world applications in resource-constrained environments. The expected outcome of this project is an innovative image classification method that competes with CNN models while offering lower energy consumption and easier deployment across various industrial and scientific domains. This research seeks to provide a revolutionary alternative to GPUs, reducing costs and energy consumption in computer vision.