<p>Internal browning severely compromises the quality and storage efficiency of pears, creating an urgent need for rapid and accurate detection. A method combining acoustic-vibration non-destructive testing technology with an improved MobileViT model was proposed to enhance the accuracy of distinguishing between slightly and severely browned pears. First, acoustic-vibration signals were collected from fruits using a self-developed pear-acoustic-vibration detection system. Second, the Symmetrized Dot Pattern (SDP) method was employed to convert the one-dimensional acoustic-vibration signal into a two-dimensional “snowflake pattern” image. Subsequently, the Symmetry Consistency Branch (SCB) and Polar Structure Attention (PSA) modules were introduced to enhance the MobileViT model’s ability to capture both local and global information within the SDP image. Experimental results indicate that when the time interval coefficient <i>t</i> is 25, the magnification angle <i>ζ</i> is 60°, and the rotation angle <i>θ</i> is 60°, the SDP images of samples with different browning degrees show the most significant differences without overlap. Under these conditions, the improved MobileViT model achieves a classification accuracy of 93.3% on an independent test set comprising 150 sample images (healthy pears <i>N</i> = 84, slightly browned pears <i>N</i> = 48, severely browned pears <i>N</i> = 18), demonstrating high discriminative capability even for slightly browning pears (91.7%). Comparative analysis revealed that the improved MobileViT model significantly outperformed mainstream architectures (DenseNet, VGG16, and MobileNetV3), achieving macro-average recall, macro-average precision, and <i>F</i>₁-scores of 93.39%, 92.93%, and 93.14%, respectively, on the test dataset. These findings provide valuable insights for developing online detection and automated grading technologies for early-stage internal browning in pears using acoustic-vibration methods.</p> Graphical Abstract <p></p>