Genetic Programming for Image Classification and Feature Learning

Genetic programming (GP) applies evolutionary principles to automatically evolve computer programs that perform image classification and feature learning. Candidate solutions are represented as tree-structured combinations of terminals (data inputs) and functions (operators), and populations of such trees undergo selection, crossover and mutation guided by a fitness function. This process yields bespoke image-processing pipelines that extract and combine informative features without manual engineering. GP has demonstrated the ability to evolve unified programmes that both localise and classify multiple object categories, construct novel representations from raw pixels and adapt to varied domains with limited data. Hybrid frameworks that integrate GP with deep learning modules have recently emerged, leveraging the interpretability and flexibility of GP alongside the representation power of neural networks. Applications span medical diagnostics, remote sensing, industrial inspection and autonomous systems, where GP’s evolved models offer transparent decision rules, reduced parameter counts and competitive performance under data-scarce conditions.

Research from Nature Portfolio

No recent Nature Portfolio content available.

Research from all publishers

One foundational study introduced a domain-independent window approach for multiclass object detection using GP. By defining terminal sets based on pixel statistics and selecting a minimal function set, the evolved programme scans large images to locate and classify diverse object classes, optimising detection rate while controlling false alarms. This work demonstrated a single GP individual could simultaneously produce classification maps and localisation outputs, highlighting GP’s versatility in object detection and multiclass classification. A more recent investigation applied GP to estimate liver damage from high-resolution medical images. This approach incorporated image preprocessing, colour-space transformations and a customised GP-driven regression pipeline with hyperparameter tuning. The evolved models achieved R² values above 0.5 and low mean squared errors, outperforming methods that lacked preprocessing. The study underscored GP’s capability to learn meaningful features automatically and configure regression models that rival invasive assessment techniques.

Genetic Programming for Image Classification and Feature Learning publication trend

The graph below shows the total number of articles in genetic programming for image classification and feature learning across all publications each year (not limited to Nature Index journals).

Technical terms

Genetic Programming: An evolutionary algorithm that evolves computer programmes by simulating natural selection, aiming to solve tasks such as image classification.
Terminal Set: The inputs or operands in a GP individual, typically representing raw data or extracted features.
Function Set: The operations or primitives in GP that combine terminal inputs to form program logic.
Fitness Function: A quantitative measure used to evaluate and rank GP individuals based on task performance.
Feature Learning: The process by which algorithms automatically discover and optimise data representations suitable for classification or regression.

References

1. A Domain-Independent Window Approach to Multiclass Object Detection Using Genetic Programming. EURASIP Journal on Advances in Signal Processing (2003).
2. Imaging Estimation for Liver Damage Using Automated Approach Based on Genetic Programming. Mathematical and Computational Applications (2025).