Code smells are undesirable design characteristics in software source code that negatively affect maintainability, readability, and scalability. By detecting code smells, it becomes possible to identify code segments requiring refactoring, thus enhancing code quality. Although machine learning techniques have been widely applied in previous studies for code smell detection, challenges persist regarding detection accuracy and generalization capability. In this study, we propose two models: a directly isomorphic ensemble model combining convolutional neural network (CNN) and a random forest-negative correlation learning-CNN (RNCNN), to detect four types of code smells: Data class, God class, Feature envy, and Long method. A feature selection technique based on random forests is employed to extract the most informative features, and deep learning techniques are utilized to model the structural characteristics of the code. The model is then trained using a homogeneous ensemble deep learning approach based on negative correlation learning. Experimental results demonstrate that the proposed method improves the F1 score by up to 4.93% compared to baseline approaches.