海华AI比赛感受 Haihua AI garbage classification competition
看到垃圾分类这个标题时我就决定要参加这个比赛。我之前也参加过类似的项目课题研究，只是并不专业，更注重社科题材的研究。那时我请教老师，老师说用机器学习构建神经网络然后图像识别进行垃圾分类其实是个很简单的事情。于是我变很好奇到底是怎么做到的。
What was it? I decided to take part in Haihua Al rubbish classification competition when I saw the title ‘rubbish classification’, which is to use computer classify the rubbish. After all, computer science is the major I have chosen. At that time, I asked my teacher for some advice, and he said that it was actually just using the machine learning to build neural networks and then use image recognition to classify garbage. Then, I truly became curious about how it could be done.
参加比赛前，我们进行了学习。老师从如何预测简单的商店营业额一直教到图像处理，看似无关的知识点之间环环相扣，最终实现了创建模型，识别图片并垃圾分类。刚学完时，我都不敢相信课程已经结束了，原来创建模型真的不是一件难事。安装配备tensorflow等包（也可以尝试其他环境），导入数据和图片，用图片信息进行编号，分成训练集和测试集，合理运用Keras官网上准确度较高的模型，训练自己的模型，最终运行模型获得预测数据，整过程便结束了。
My group learnt some relevant knowledge before we took part in the competition, being taught from how to predict the turnover of a store by plotting graph to how to do image processing, from simple to difficult. Seemingly irrelevant knowledge points were interlinked, such as which PS skills can be realized by using the image processing codes, and finally the model, which was suitable for the competition, was created and images were recognized and rubbish classification was realized. When I finished, I couldn’t believe the course was over, because there were too many things for me to learn that the time really flew. It turned out that it wasn’t so hard to create models. Installing packages such as tensorflow (you can also try other environments), importing data and pictures, numbering them with picture information …… and finally running the model to obtain the prediction data, the whole process is over.
别看只写了两三行字，这中间可是用了几周的时间，训练模型确实是件苦差事。从开始二三十遍为一循环，到十次一循环，到最后一次一循环，准确度始终在99.7徘徊，却始终没有更好的成绩，确实是一件令人着急的事。还有上学的缘故，家里电脑硬件不足，我还学会了远程操控电脑。其中还有不少困难：电脑硬件出现问题，电脑无法连接，硬盘内存满了要换一个盘，环境未匹配上，下载官网模型时间过长通过其他办法下载……虽然有困难，但我们都一一解决了。这也有助于我对这一块知识的了解，比如在学习编程之前，我可以说是对电脑一无所知，只会上网。而现在，不论是编程语言的更加熟练，还是对电脑的更加熟悉，都有助于我将来的学习，毕竟这是我选定的专业方向。当然，经历了这么多，我也逐渐能做到做事时能不像以前那么着急，不能图快，一件一件慢慢处理才能达到最高的效率。
Despite writing only two or three lines of codes, it took weeks to train the model. From the beginning with 20 or 30 times for a cycle of training, to 10 times a cycle, to at last once a cycle, the accuracy was always hovering in 99.7 percent. But that there is no more better accuracy is really a matter of urgency. Also because of school schedule and the insufficient home computer hardware, I also learned to operate the computer remotely, which can save my time on the way. Meanwhile, there were also many difficulties, such as computer hardware problems, Internet connection difficulties, the hard disk memory is full, the environment is not matched, and so on. Although there were difficulties, we solved them one by one which added to my deeper understanding of the programme. For instance, before learning programming, I can say that I knew nothing about computers, except the Internet. Now,being more proficient in the programming language and more familiar with the computer can absolutely help me in my future studies. By taking part in this competition I learn a lot of important skills such as time management and working in a methodical and reflective way.
曾经，我觉得这是一个多么遥远的专业，因为不了解，所以觉得简直是不可能的事情。但如今回头再看，其实没有我想的那么难。学习就是这样，是一个上升的阶梯，从下往上看时感觉遥不可及，但当不知不觉爬上去后往下看，似乎也没那么遥远。
In the past, I thought learning this programme would be beyond my capabilities because I was not familiar with it and it felt impossible for me to understand. However, looking back now, it was not as hard as I thought. Learning is just like this, a ladder: when I get from the bottom to the top, I feel it far away; when unwittingly climbing up and looking down, it seems not so far away.