The detection of unsound wheat kernels in traditional wheat purchasing is affected by human factors, resulting in wrong wheat grading. At present, computer-based recognition of wheat kernels has generally low accuracy, and few types of wheat kernels can be recognized. To quickly, accurately, and objectively recognize wheat kernels, this study proposed an improved strategy of wheat kernel recognition method based on deep learning. First, a large number of collected wheat images were labeled, and the wheat kernels were divided into five categories: perfect kernels, broken kernels, impurities, sprouted kernels, and moldy kernels. Second, the improved strategies of VggNet-16, ResNet-34, EfficientNet-b2, DenseNet121, and Vit models were proposed. Based on the two-stage target detection method, the improved network model was used to detect wheat kernels. Moreover, the accuracy of the model was verified by performing comparative tests. Results show that the improved network structure is obviously improved, and the highest accuracy rate of wheat kernel identification is 96%. The precision, recall rate, and F1-score of VggNet-16-W, ResNet-34-W, EfficientNet-b2-W, and DenseNet121-W models are above 97%. This study provides a good reference for rapid and accurate detection of wheat quality. Keywords: deep learning; image recognition; improved strategies; network model; wheat kernels

Person re-identification (ReID) has attracted the attention of a large number of researchers due to its wide range of applications. However, due to the difficulty of extracting robust features and the complexity of the feature extraction process, ReID is difficult to truly apply in practice. In this paper, we utilize Pyramid Vision Transformer (PVT) as the backbone for feature extraction and propose a PVT-based ReID method in conjunction with other studies. First, we establish a basic model using powerful methods verified on CNN-based ReID. Second, to further improve the robustness of the features extracted from the PVT backbone, we design two new modules: (1) a local feature clustering (LFC) module is used to select the most discrete local features and cluster them individually by calculating the distance between local and global features, and (2) side information embeddings (SIE) are used to encode nonvisual information and send it to the network for use training in order to reduce its impact on the features. Our experiments show that the proposed PVTReID achieves an mAP of 63.2% on MSMT17 and 80.5% on DukeMTMC-reID. In addition, we evaluated the inference speed for images achieved by different methods, proving that image inference is faster with our proposed method. These results clearly illustrate that using PVT as a backbone network with LFC and SIE modules can improve inference speed while extracting robust features.

Due to the influence of background conditions, lighting conditions, occlusion issues and the image resolution, how to extract robust person features is one of the difficulties in ReID research. Vision in Transformers (ViT) has achieved significant results in the field of computer vision. However, the existing problems still limit its application in ReID due to slow extraction of person features and difficulty in utilizing local features of people. To solve the mentioned problems, we utilize Pyramid Vision Transformer (PVT) as the backbone of feature extraction and propose a PVT-based ReID method in conjunction with other studies. Firstly, some improvements suitable for ReID are used on the PVT backbone, and we establish a basic model by using powerful methods verified on CNN-based ReID. Secondly, in an effort to further promote the robustness of the person features extracted by the PVT backbone, two new modules are designed. (1) The local feature clustering (LFC) is recommend to enhance the robustness of person features by calculating the distance between local features and global feature to select the most discrete local features and clustering them. (2) The side information embeddings (SIE) are used to encode non-visual information and send it into the network for training to reduce its impact on person features. Finally, the experiments show that PVTReID has achieved excellent results in ReID datasets and are 20% faster on average than CNN-based ReID methods.