摘要：【中文】本发明提供一种基于人工智能的手机拍照识别垃圾分类方法，用户在手机移动端进行拍照垃圾识别分类时，首先由用户选择垃圾分类标准，分类标准可以是城市名称，也可以是其他的分类标准，用户选择的分类标准将被记录下，之后用户可以点击屏幕上的采集图像按钮来选择要识别的图像，识别图像后，图像上的垃圾位置和分类结果等信息将会标注在原图像上，使用户了解到图像上垃圾的位置及其相应的分类信息，最后用户可以选择是否保存图片结果，如果选择保存，则将处理后的图片保存在手机并发送到服务器端。本方法大概率对垃圾进行识别并正确分类，提高了分类的准确性。 【EN】The invention provides a mobile phone photographing and garbage recognition classification method based on artificial intelligence, when a user performs photographing garbage recognition classification at a mobile terminal of a mobile phone, the user selects a garbage classification standard, the classification standard can be a city name or other classification standards, the classification standard selected by the user is recorded, the user can click an image acquisition button on a screen to select an image to be recognized, after the image is recognized, information such as garbage position and classification result on the image is marked on an original image, so that the user can know the garbage position on the image and corresponding classification information thereof, and finally the user can select whether to save the image result, if the image is selected to be saved, the processed image is saved in the mobile phone and is sent to a server terminal. The method identifies and correctly classifies the garbage with high probability, and improves the classification accuracy.

摘要：【中文】本发明提供了一种跨平台在移动端动态接收人工智能模型训练通知的方法，在人工智能技术开发中，大量深度学习模型任务需要在云端使用GPU集群完成训练任务，通过使用移动端设备可以实时查看深度学习模型训练状态消息通知，将会大大提升用户查看模型训练状态、实时查看训练状态反馈的体验，参与竞赛的过程越来越便利、高效。为此，本发明提出的技术方案，能够实现在移动端HTML5页面或本地端PC等设备上将竞赛训练任务指令发送到GPU服务器进行训练，在移动端设备中随时查看训练通知、训练过程的方法，从而解决了用户在进行竞赛训练的过程中查看训练过程和通知的单一使用场景限制。用户能够在任意场景下通过移动端设备实时查看竞赛训练过程和结果通知，提升了参与竞赛训练的用户体验和竞赛训练时间效率。 【EN】The invention provides a method for dynamically receiving an artificial intelligence model training notice at a mobile terminal across platforms, in the development of an artificial intelligence technology, a large number of deep learning model tasks need to use a GPU cluster at the cloud to complete the training tasks, the deep learning model training state message notice can be checked in real time by using mobile terminal equipment, the experience of checking model training states and checking training state feedback in real time by a user can be greatly improved, and the process of participating in a competition is more and more convenient and efficient. Therefore, the technical scheme provided by the invention can realize the method that the competition training task instruction is sent to the GPU server for training on the equipment such as the mobile end HTML5 page or the local end PC and the like, and the training notice and the training process are checked at any time in the mobile end equipment, thereby solving the problem that the user checks the training process and the single use scene limit of the notice in the competition training process. The user can check the competition training process and result notification in real time through the mobile terminal device under any scene, and the user experience and the competition training time efficiency of the competition training are improved.

摘要：【中文】本发明提供一种基于人工智能的垃圾分类方法，包括：第一步，划分数据集；读取垃圾数据集标注文件，所述垃圾数据集标注文件中包括预先设置的垃圾图片路径和该垃圾图片所对应的垃圾类别；将所有数据按照一定比例划分为训练集、验证集和测试集；第二步，确定循环学习率中学习率的最大值和最小值；第三步，通过确定了的循环学习率的最大值和最小值，通过循环学习率函数来得到当前模型优化的学习率，进而优化模型参数；第四步，模型测试；加载上述模型和参数，来对测试的垃圾图片进行类别预测，完成垃圾分类。 【EN】The invention provides a garbage classification method based on artificial intelligence, which comprises the following steps: the first step, dividing a data set; reading a junk data set marking file, wherein the junk data set marking file comprises a preset junk picture path and a junk category corresponding to the junk picture; dividing all data into a training set, a verification set and a test set according to a certain proportion; secondly, determining the maximum value and the minimum value of the learning rate in the cyclic learning rate; thirdly, obtaining the optimized learning rate of the current model through a cyclic learning rate function according to the determined maximum value and the determined minimum value of the cyclic learning rate, and further optimizing the model parameters; fourthly, testing the model; and loading the model and the parameters to predict the category of the tested junk pictures so as to finish the junk classification.

摘要：【中文】本发明公开了一种大型电渣炉高强度冷却定向凝固结晶器及凝固工艺，属于电渣特种冶金技术领域，解决了现有技术中超大直径电渣锭现有的凝固条件恶化导致的凝固方向改变和易出现疏松缩孔等问题。本发明结晶器为分段式组装结晶器，设置有多段结晶器单元，结晶器单元为可拆卸的；各段结晶器单元的内径均从下往上均匀增加，相邻两结晶器单元中，上方的结晶器单元底端内径与下方的结晶器单元顶端内径相同，各段结晶器单元内表面斜度一致，斜度α≤10°。本发明结晶器及凝固工艺适用于直径超过1600mm或吨位超过60吨的大型电渣锭的熔炼。 【EN】The invention discloses a high-strength cooling directional solidification crystallizer of a large electroslag furnace and a solidification process, belongs to the technical field of electroslag special metallurgy, and solves the problems that the solidification direction is changed and loose shrinkage cavities are easy to appear due to the deterioration of the existing solidification conditions of an electroslag ingot with an overlarge diameter in the prior art.

摘要：【中文】本发明提出一种基于用户数据的全自动竞赛数据处理系统以及方法。该技术方案主要包括自动化本地启动文件版本检测、自动化本地环境安装、提交代码以及自动化提交代码三大部分。在用户操作使用上操作简单，只需要两步就能实现本地环境自动安装并运行；在程序运行上，实现环境自动版本校验，保证本地启动文件是最新文件，并且用户自定义配置操作简单，能通过用户代码自动获取用户代码中所需要安装的包名，从而通过精准的差异化对比，获得更为准确的用户安装需求；配置精准的历史记录差异化对比，更为精准的避免重复安装，节约用户安装时间；在精准获取需要安装包名后，进行批量安装，安装速度更快；保留历史记录，有效避免重复安装。 【EN】The invention provides a full-automatic competition data processing system and method based on user data. The technical scheme mainly comprises three parts, namely automatic local starting file version detection, automatic local environment installation, code submission and automatic code submission. The operation is simple in the aspect of user operation, and the automatic installation and operation of the local environment can be realized only by two steps; in the process of program operation, automatic version verification of the environment is realized, the local boot file is ensured to be the newest file, the user-defined configuration operation is simple, and the package name required to be installed in the user code can be automatically obtained through the user code, so that more accurate user installation requirements can be obtained through accurate differential comparison; the differential comparison of the configured accurate historical records can avoid repeated installation more accurately, and the installation time of a user is saved; after the package names needing to be installed are accurately obtained, batch installation is carried out, and the installation speed is higher; and history records are kept, and repeated installation is effectively avoided.