In this paper, a new quantum convolutional neural network (QCNN) design with a quantum wavelet transform (QWT) is presented in order to address the limitation of classical CNNs dealing with high-dimensional noisy images. Capitalizing on quantum principles such as superposition, entanglement, and parallelism, the new model enables efficient multi-resolution analysis. QWT, implemented with parameterized quantum circuits (PQCs) consisting of Hadamard gates, controlled rotations, and lifting operators, decompose quantum-encoded images into high-frequency (texture, edges) and low-frequency (structural) components. Depthwise separable quantum convolutions (QuC) to reduce computational complexity from \(\mathcal{O}\left({n}^{2}\right)\) to \(\mathcal{O}\left(n\right)\) , entanglement-based adaptive quantum pooling to preserve spatial correlations, and quantum batch normalization (QBN) to stabilize training without collapsing superposition are key breakthroughs. To encourage resilience in noisy intermediate-scale quantum (NISQ) hardware, the architecture is designed with topological quantum error correction (QEC) and achieves 98% fidelity in feature maps with 5% gate error. Experimental results on a range of datasets—MNIST (99% accuracy), FashionMNIST (82%), high-resolution medical liver images (95.4% tumor detection rate), and Pepper images (94.1% object classification)—demonstrate superior performance to conventional CNNs. 5-ms-per-image model operates on IBM's 65-qubit quantum processor and can scale and be made feasible in real-world applications related to medical imaging and satellite imaging. The research fills in the gap between traditional image processing and quantum processing and offers a noise-robust, efficient resource model to handle intricate visual information.
