Authors :

binti Sumari and Norisma Halim

Address :

School of Computer Science, Taylor’s University Malaysia

School of Computer Sciences, Universiti Sains Malaysia Pulau Pinang, malaysia

Abstract :

Geological pattern recognition is essential for interpreting subsurface structures, classifying lithological units, and guiding exploration activities. This study presents a novel framework that leverages Morphological Feature Learning integrated with Tree Seed Optimization for enhanced geological pattern recognition. Traditional methods often suffer from low accuracy in complex geological environments due to inadequate feature extraction and suboptimal parameter tuning. To address these challenges, we propose a Morphological Convolutional Neural Network (MorphCNN) that embeds morphological operations such as dilation and erosion into convolutional layers, enabling better extraction of shape and texture features relevant to geological formations. Tree Seed Optimization (TSO) is employed to automatically fine-tune hyperparameters, boosting the model's performance and convergence speed. The proposed framework is applied to lithofacies classification in seismic images, where it effectively captures structural features and distinguishes between different geological units. Experimental results show a significant improvement in classification accuracy, robustness to noise, and interpretability of learned features compared to conventional CNNs. This confirms the effectiveness of integrating morphological priors and bioinspired optimization in geological pattern recognition. The proposed method gradually improves the filter size by 94%, learning rate by 96.2%, batch size by 90%, and number of tree seeds by 95.7%.

Keywords :

Geological Pattern Recognition, Morphological Feature Learning, Tree Seed Optimization, Morph-CNN, Lithofacies Classification, Seismic Image Analysis.

DOI :