Image classification is a crucial task in machine learning. In recent years, this field has witnessed rapid development, with a series of image classification models being proposed and achieving state-of-the-art (SOTA) results. Parallelly, with the advancement of quantum technologies, quantum machine learning has attracted a lot of interest. In particular, a class of algorithms known as variational quantum algorithms (VQAs) has been extensively studied to improve the performance of classical machine learning. In this paper, we propose a novel image classification framework using VQAs. The major advantage of our framework is the elimination of the need for the global pooling operation typically performed at the end of classical image classification models. While global pooling can help to reduce computational complexity, it often results in a significant loss of information. By removing the global pooling module before the output layer, our approach allows for effectively capturing more discriminative features and fine-grained details in images, leading to improved classification performance. Moreover, employing VQAs enables our framework to have fewer parameters compared to the classical framework, even in the absence of global pooling, which makes it more advantageous in preventing overfitting. We apply our method to different SOTA image classification models and demonstrate the superiority of the proposed quantum architecture over its classical counterpart through a series of experiments on public datasets.