Lung cancer is the leading cause of cancer mortality worldwide. Identifying the subtypes of lung cancer accurately is crucial for the diagnosis and treatment of lung lesions in clinical practice. Artificial intelligence (AI) methods employed for lung cancer classification mainly rely on CT images, histopathological images, and a few methods based on surgical lesion section images. However, there is a great need for methods to utilize surgical lesion section images of lung tumors, a type of non-standard medical images, to assist surgeons in rapidly deciding surgical plans during the operation. This work proposes an attention-based classification network named LungNeXt for surgical lesion section images of lung tumors. This work integrates the ultra-coordinate attention (UCA) module into the ConvNeXt model to enhance the focus on the critical lesion region within the image. Further, this work embeds the self-attention (SA) module to capture the global long-range contextual dependencies within the image. Besides, this work employs Real-ESRGAN algorithm for image pre-processing to enhance the resolution and reduce the noise of the images. To alleviate the challenge of inherent class imbalance, this work utilizes class-balanced loss during training. LungNeXt achieves outstanding performance with 89.39% accuracy and 0.9638 AUC, highlight its potential as a reference for surgeons in rapidly deciding surgical plans and treatment strategies.