<p>Quantum Neural Networks (QNNs) present a promising research direction in image recognition. However, models that rely on a single circuit topology frequently exhibit limited expressive capacity, thereby constraining scalability and recognition accuracy. To address these limitations, we propose a Hybrid Quantum–Classical Inception-Inspired Convolutional Neural Networks (HQICNN). The specific descriptions are as follows. First, in the encoding layer, amplitude encoding layer is adopted to map complete multi-channel images data into quantum states. Second, in the convolutional layer, this architecture utilizes quantum convolutional kernels with heterogeneous circuit topologies for feature extraction. A Hadamard integration method is proposed to aggregate multi-channel feature information. Then, in the pooling layer, quantum pooling using RY gates is employed to reduce feature dimensionality. Finally, in the prediction layer, a fully connected layer is applied to map the quantum-enhancing features into class labels for classification tasks. The quantum resource analysis indicates that for the MNIST-4 classification task, the quantum circuit depth of the proposed HQICNN is approximately <i>O</i>(<i>n</i>), which is lower to the existing HQCNN model(<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(O(n^2)\)</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mi>O</mi> <mo stretchy="false">(</mo> <msup> <mi>n</mi> <mn>2</mn> </msup> <mo stretchy="false">)</mo> </mrow> </math></EquationSource> </InlineEquation>). Experimental results demonstrate that HQICNN achieves 98.3% accuracy on the MNIST-4 classification task, outperforming existing HQCNN models (90.0%). Furthermore, the HQICNN model utilizes only 11% of the parameters of a CNN while achieving comparable classification performance. The proposed design enhances image classification accuracy within a quantum circuit architecture while reducing the number of parameters, performing better than HQCNN models.</p>