With the improvement of the level of information technology, human-computer interaction technology began to gradually come into people's view, and constantly penetrate and develop in a diversified direction. Based on the research hotspot of somatic interaction in human-computer interaction technology, this paper combines action recognition with cultural digital interaction experience. This paper, based on the research hotspot of somatosensory interaction in human-computer interaction technology, we propose a pose-action recognition technology scheme based on AdaBoost algorithm to optimise classifiers for machine learning training in order to solve the efficiency problem of human-computer interaction.

Grapes are common fruits in life, but grape varieties are diverse, mainly by the naked eye to distinguish, in order to solve the problem of low efficiency of artificial differentiation, using deep learning to achieve classification of a variety of grape varieties, constructing a CNN-GLS model, taking five kinds of grape leaf images as the research object. By learning and extracting the feature information of grape leaves through multi-layer convolutional network, adding a data enhancement layer to the network to reduce model overfitting, using the Leaky Relu activation function to avoid aliasing in the gradient direction, adding flat layer Flatten "flattening" convolutional pooled data, and one-dimensional multi-dimensional data, Adam optimizer can improve sparse gradient performance, reduce training time, and improve network efficiency. The experimental results show that the accuracy of small sample grape leaf species recognition based on convolutional neural network CNN-GLS model is 95.13%, and the model has robustness and high generalization ability, which is of great significance for the taxonomic identification of plants.

If a crop has a disease, it will lead to a significant decline in crop yield, which will also affect the quality of crops, so it is necessary to identify diseases and treat them in time. Deep learning technology is used to identify diseases, leaf images of 61 crop diseases are used as research objects, disease features with strong expression ability of images are extracted by inception_ v3 network until the bottleneck layer, stored in the bottleneck layer file, used as input of the subsequent network layer, the training parameters are adjusted, the weight is updated by gradient descent method, overfitting is alleviated, and the image is classified and recognized by softmax. Considering the practical application, a server for testing crop diseases based on a training model based on Inception_v3 network combined with transfer learning was developed. Integrate the front and back ends, load the images to be tested on the server test page, and the results show the disease category. The experimental results show that the accuracy of Inception_v3 neural network combined with transfer learning strategy for identifying crop disease species is 93.90%, and the model has robustness and high generalization ability, which is of great significance for the study of crop disease identification.

With the development of blockchain, and its security issues have attracted the attention of more and more researchers. Currently, the related research is still in its infancy. This paper first reviews the basic technology and development of blockchain, summarized and introduced the blockchain structure. In terms of blockchain security, we analyze and summarizes the domestic and foreign literatures in recent years, and divide blockchain security issues into three aspects: protocol security, privacy security and system security, and the problems of blockchain security in these three aspects are analyzed. At present, protocol security research mainly focuses on the encryption mechanism of blockchain technology and smart contract vulnerabilities; Privacy security research covering mainly the potential security issues facing blockchain applications; The system security research mainly includes the classification of attack methods on blockchain. Then, we summarized the defense strategies to deal with the above security issues and analyzed the shortcomings and problems. Finally, based on the current research status of blockchain, we point out two future research directions.

支持向量机的核函数的应用性越来越强.性能优秀的核函数可以带来非常好的效果,为了充分利用核函数的优点,构造出一种创新的多核集成方法.在这个整体框架中,每个核回归器都与一个权重相关联,该权重可以根据其对回归结果的贡献来自训练自动调整.通过这种方式,可以直接从数据中学习更合适的核函数类别及其对应的参数,而无需任何人工干预,从而有更好的回归性能.同时为了使非凸问题可以求解,将引入L1范数和L2范数进行重新建模,从而获得解耦形式,具有可求偏导形式的模型.在一些UCI回归数据集上的实验结果表明,论文提出的方法在最新的比较方法中获得了最佳的回归性能.

To improve the recognition rate and stability of weed identification for single features, A support vector machine based multiple kernel ensemble method for weed identification is proposed The three weed features are trained separately to derive a pool of kernel functions., multiple kernel classification results of the three single features are used as independent evidence to construct the basic probability assignment, introduce the complexity of the weed recognition algorithm based on multi-kernel non-convex optimization, and give the final recognition results according to the ensemble results and the classification judgment threshold. The experimental results show that the recognition rate of the multi-kernel non-convex optimized weed identification method reaches 97.57

为揭示耕作方式对小麦根系构型及生长趋势影响的作用机理.基于根系构型数字化平台研究全幅旋耕(FT)和2种带状耕作(ST1、ST2)方式对小麦根系3D构型及生长趋势的影响,提出了土体空间覆盖率差异系数(P)、角度拓展趋势差异系数(Pa)、长度拓展趋势差异系数(PL)3种量化指标.结果表明,带状耕作条件下,小麦根系沿种沟方向生长趋势明显,ST1处理下在距离苗期14、28 d时沿种沟方向土体P、Pa、PL分别比FT处理高出1455％、4520％、3583％和2610％、241％、740％.ST1、ST2处理相比较发现,涂抹后的种沟壁对根系生长的限制更为明显,在距离苗期14、28 d时,ST1处理下小麦根系沿种沟方向P、Pa、PL分别比ST2处理高出78％、88％、31％和20％、81％、37％.耕作方式直接影响小麦根系的空间拓展趋势,本研究提出的研究方法和评价指标能够简明耕作方式对小麦根系生长趋势影响的作用机理,为耕作机具的优化设计提供更加全面的理论支持.

Compared to the previous use of machine learning algorithms for recognition, its disadvantages have small accuracy recognition speed is slow. In this paper, we extend the idea of weed recognition to deep convolutional neural networks, and after effective training can get a network model with relatively good robustness, which can provide great help for weeding in corn fields. We choose the deep convolutional model for comparison with the common weeds in corn fields as the dataset, and we can get that the VGG16 model has better recognition effect on weeds compared with other convolutional models through experiments. Through multiple rounds of training, the F1-values of VGG16 model for Black-grass, Charlock, Cleavers, Common chickweed, and Loose silky-bent were 0.971, 0.945, 0.949, 0.959, and 0.958, respectively, and we selected different optimizers for the VGG16 model to retrain these weed datasets to obtain the SGD optimizer with the The combined performance is the best. In summary, by combining deep learning in machine vision to weed identification, we can effectively reduce the labor cost in agricultural farming as well as can better increase the yield of corn.

科研项目包括国家各级政府成立基金支撑的纵向科研项目、来自企事业单位的横向科研合作开发项目和学院自筹科研项目等,无论哪一类课题,均是对系列独特的、复杂的、相互关联的活动进行研究,老师和学生的协作研究能够培养学生的创新思维和独立工作能力.高校教师个人或团队可以根据各自涉及的科研项目内容,采用文献研究法、统计分析法等方法,进行以科研项目为导向的教育创新与实践研究,有助于不断优化和完善高职院校创新创业教育理论体系.

•We develop a novel regression method based on kernel trick and ensemble principle. Its merit is that multi-kernel selection and parameter decision can be conducted automatically through a pool of kernels.•In our proposed method, we introduce sparsity to evaluate the quality of the model. With this sparsity model, well-behaved regressors are selected and the impacts of badly-behaved regressors are decreased.•Experimental results on UCI regression and computer vision datasets indicate that compared to other regression ensemble methods, such as random forest and XGBoost, our method has the advantages of best performances in keeping lowest regression loss and highest classification accuracy.

In view of the non-linear change of face image caused by illumination, the recognition rate is reduced and the ability of single feature expression is limited. Based on local binary mode (LBP) and Gabor wavelet, we propose a face recognition method. By extracting these two local features which are robust to illumination, we use generalized discriminant analysis (GDA) to reduce the dimension and then use discriminant correlation analysis (DCA) to fuse the features. Experiments on the ORL, Extended YaleB, OFD, and CAS-PEAL face databases demonstrate that the proposed method works better than LBP or Gabor alone.