Research on a highly efficient and accurate detection algorithm for bamboo strip defects based on deep learning

Surface defects on bamboo strips significantly impact the appearance quality and mechanical strength of bamboo laminated timber. Traditional manual methods for detecting surface defects on bamboo strips are inefficient, subjective, and lack standardization, resulting in misjudgments, missed detections, and inconsistent outcomes, which fail to meet modern industrial demands. To address this, the study proposes a target detection algorithm for efficiently and accurately detecting bamboo strip defects. The algorithm is based on a diverse dataset of 10 defect types and 6523 images, built on the YOLOv8 benchmark model and incorporated the DySample module, SPPFUniRepLKA module, and EIoU loss function to create four bamboo strip defect detection models: Ourwork (n, s, m, l). The results demonstrate that the Ourwork-n model achieves an optimal balance between performance and complexity, with a mAP@0. 5 of 96. 5 %, a mAP@0. 50: 0. 95 of 71. 6 %, Precision of 94. 1 %, Recall of 92. 6 %, and an F1 score of 93. 3 %. These improvements correspond to increases of 1. 1 %, 1. 8 %, 0. 9 %, 1. 3 %, and 1. 1 %, respectively, compared with the YOLOv8 benchmark model. The Ourwork-n model can meet industrial detection requirements with both high accuracy and good real-time performance (42 Frames Per Second), providing an effective solution for the efficient and precise detection of bamboo strip defects, and ensuring the high-quality production of bamboo laminated timber. • A perspective transformation method is proposed to reduce training instability. • The baseline model is enhanced by improving three critical architecture modules. • Four model sizes are developed for different industrial application scenarios. • The method is applied to quality control in bamboo laminated timber production. • The small model achieves 96. 5 % precision and 42 frames per second in detection.