Research and Application of Lightweight Network Architecture for Real‐Time Detection of TFT‐LCD Display Defects

The mura defects of thin‐film transistor liquid crystal display (TFT‐LCD) panels exhibit characteristics of low contrast and random positioning, resulting in issues such as low accuracy in defect identification and poor user experience. This paper proposes a lightweight YOLO‐SPPAM network. Based on the spatial pyramid pooling (SPP) module belonging to YOLOX, the newly constructed spatial pyramid pooling attention (SPPA) module allows the network to focus on salient target regions, enhancing the model's ability to perceive crucial features. This paper introduces augmentable convolutional block attention module (ACBAM) to obtain parallel dual‐channel attention by parallel processing of channel attention and spatial attention. The paper replaces ordinary convolutions in down sampling with fine‐grained separable convolution module (FGSCM). Qualitative and quantitative comparison experiments with state‐of‐the‐art algorithms on a self‐made TFT‐LCD mura defects dataset demonstrate that YOLO‐SPPAM outperforms in terms of accuracy and speed, meeting the real‐time requirements of TFT‐LCD defect detection tasks.