基于木材纹理图像和改进ResNet50_DTPE模型的5种红木树种识别方法OA北大核心CSTPCD

In order to identify five Hongmu species,Pterocarpus erinaceus Poir.,Pterocarpus macrocarpus Kurz,Dalbergia cochinchinensis Pierre,Dalbergia latifolia Roxb.and Dalbergia retusa Hemsl.,a new recognition model for wood texture was proposed.This model aims to provide a fast,simple,efficient and accurate method,based on the improved attention mechanism algorithm of residual network(ResNet50)and a detailed texture perception engine(DTPE).The images of wood patterns on the surface of Hongmu furniture were photographed,and the wood texture images were extracted to build the dataset.Stable diffusion combined with control network technology was used to generate expanded image dataset.At the same time,data enhancement processing was performed to construct ResNet50_DTPE model,which was then compared with the existing ConvNeXt and EfficientNetV2 network models.The results indicate that dataset expansion and enhancement increase the number of images and enhance the image quality,positively impacting the model performance.DTPE is integrated at the end of ResNet50 to maintain the original network structure of ResNet50,resulting in high recognition accuracy.The ablation test results verified the effectiveness of ResNet50_DTPE model.The training time was 11.0 hours,the inference time was 24.0 ms/sheet,and the highest recognition accuracy was 99.8％,indicating that ResNet50_DTPE is feasible for identifying the five Hongmu species.