ISDA-QBAS Net: A Novel CNN Optimization and Data Enhancement Model for Pneumonia Diagnosis

Convolutional neural networks (CNNs) excel in medical image analysis through iterative weight learning and updating. Traditional random weight initialization can result in low accuracy, prolonged convergence, or training failure due to extreme parameter values. This paper introduces the Quantum Beetle Antennae Search (QBAS) algorithm, a bionic optimization method. To address the limitations of the Chest X-ray (CXR) dataset, the Implicit Semantic Data Enhancement (ISDA) algorithm is proposed, which semantically augments the dataset without manual intervention. The combined QBAS-ISDA net further optimizes CNN performance. Exper-iments on the CXR dataset show the QBAS net achieves 91.3% accuracy, while the QBAS-ISDA net reaches 91.8%, surpassing the baseline's 72.6%. The QBAS algorithm also outperforms other bionic optimization methods like Beetle Antennae Search (BAS), Genetic Algorithm (GA), and Particle Swarm Optimization (PSO).