A Hybrid Feature Fusion Approach With Optimized Adaptive SVM For High-Performance Multi-Class Cervical Image Classification

Abstract

Precise and effective Hybrid Feature Extraction multi-class image classification is still an arduous problem, particularly in cases with high intra-class variability and inter-class similarity. In this paper, a new hybrid feature fusion approach combined with an Optimized Adaptive Support Vector Machine (OASVM) is proposed to improve the classification capability of complicated cervical image sets. The model that is proposed here combines manually engineered features of Gray Level Co-occurrence Matrix (GLCM), Histogram of Oriented Gradients (HOG), and Local Binary Patterns (LBP), representing both spatial gradients and texture of the images. These features are normalized and combined into a strong hybrid descriptor, and then their dimensionality is reduced using Principal Component Analysis (PCA) to remove redundancy and enhance computational efficiency.

An Adaptive SVM classifier is optimized subsequently with a hybrid kernel function that adaptive weights together linear, RBF, and sigmoid kernels. The tailored kernel expression enables the model to generalize more effectively across varied feature distributions. Grid search is also utilized to fine-tune the classifier to determine the best hyperparameters, substantially enhancing classification performance. Experimental tests prove that the designed OASVM model performs better than general-purpose classifiers such as Logistic Regression, Random Forest, and traditional SVM in accuracy, precision, recall, and F1-score. The system is able to obtain the accuracy of 95.2% on a difficult multi-class cervical image database, which proves the superiority of the hybrid feature approach and adaptive learning process. This work adds a scalable and trustworthy answer for real-world image classification tasks, especially in applications such as medical imaging, agriculture, and object detection, where accuracy and interpretability are of utmost importance.