Surface defect detection in thermoelectric cooler (TEC) components is crucial for ensuring the quality and reliability of semiconductor refrigeration devices. In this paper, we propose TECDefectNet, a hybrid deep learning model that integrates VGG16 with Squeeze-and-Excitation Networks (SENet) to improve classification performance. The VGG16 backbone extracts rich feature representations from TEC images, while SENet adaptively recalibrates channel-wise responses to highlight defect-relevant information. A softmax classifier is applied for final prediction. Extensive experiments demonstrate that TECDefectNet achieves superior performance over conventional models, with an accuracy of 88.87%, recall of 89.63%, precision of 91.24%, and an F1 score of 90.42%. Additionally, the model shows enhanced computational efficiency compared to the baseline VGG16. These results suggest that TECDefectNet is a promising solution for accurate and efficient surface defect detection in TEC components, with potential applications in industrial quality control.