Medical image segmentation acts as a foundational element of medical image analysis. Yet its accuracy is frequently limited by the scale fluctuations of anatomical targets and the intricate contextual traits inherent in medical images—including vaguely defined structural boundaries and irregular shape distributions. To tackle these constraints, we design a multi-scale feature attention network (MFA-Net), customized specifically for thyroid nodule, skin lesion, and breast lesion segmentation tasks. This network framework integrates three core components: a Bidirectional Feature Pyramid Network (Bi-FPN), a Slim-neck structure, and the Convolutional Block Attention Module (CBAM). CBAM steers the model to prioritize boundary regions while filtering out irrelevant information, which in turn enhances segmentation precision. Bi-FPN facilitates more robust fusion of multi-scale features via iterative integration of top-down and bottom-up feature maps, supported by lateral and vertical connection pathways. The Slim-neck design is constructed to simplify the network’s architecture while effectively merging multi-scale representations of both target and background areas, thus enhancing the model’s overall performance. Validation across four public datasets covering thyroid ultrasound (TNUI-2021, TN-SCUI 2020), dermoscopy (ISIC 2016), and breast ultrasound (BUSI) shows that our method outperforms state-of-the-art segmentation approaches, achieving Dice similarity coefficients of 0.955, 0.971, 0.976, and 0.846, respectively. Additionally, the model maintains a compact parameter count of just 3.05 million and delivers an extremely fast inference latency of 1.9 milliseconds—metrics that significantly outperform those of current leading segmentation techniques. In summary, the proposed framework demonstrates strong performance in thyroid, skin, and breast lesion segmentation, delivering an optimal trade-off between high accuracy and computational efficiency.

Fine-grained image recognition, as a significant branch of computer vision, has become prevalent in various applications in the real world. However, this image recognition is more challenging than general image recognition due to the highly localized and subtle differences in special parts. Up to now, many classic models, including Bilinear Convolutional Neural Networks (Bilinear CNNs), Destruction and Construction Learning (DCL), etc., have emerged to make corresponding improvements. This paper focuses on optimizing the Navigator-Teacher-Scrutinizer Network (NTS-Net). The structure of NTS-Net determines its strong ability to capture subtle information areas. However, research finds that this advantage will lead to a bottleneck of the model’s learning ability. During the training process, the loss value of the training set approaches zero prematurely, which is not conducive to later model learning. Therefore, this paper proposes the INTS-Net model, in which the Stochastic Partial Swap (SPS) method is flexibly added to the feature extractor of NTS-Net. By injecting noise into the model during training, neurons are activated in a more balanced and efficient manner. In addition, we obtain a speedup of about 4.5% in test time by fusing batch normalization and convolution. Experiments conducted on CUB-200-2011 and Stanford cars demonstrate the superiority of INTS-Net.