The Mushroom Identification And Classification Of Deep Learning Based On The IOS Platform

Mushrooms always appear on the the dinner table of which contain a variety of essential amino acids and minerals.There are many kinds of mushrooms in the world,most of them grow in the wild,except a few cultivated edible mushrooms.The eatable mushrooms in which wild grown are of medical value and taste delicious,but quite a part of them are toxic,once eaten by mistake,the poisonous mushrooms can cause life danger.What necessary to identify the toxicity of mushrooms accurately are rich experience,expertise or scientific instruments,but not for naked eyes.With the application and development of artificial intelligence technology,the use of image recognition technology can help people recognize wild mushrooms correctly,reduce the risk of eating wild mushrooms,and be more helpful for scientific researchers to carry out research in the field of wild mushrooms.Using machine learning method,a mobile phone App is designed to classify and recognize mushrooms directly.It is an effective method to classify wild mushrooms and judge their toxicity.In order to achieve the above objectives,first of all,we need to construct data sets for depth learning,for the limited condition,it is impossible to construct a complete wild mushroom image database in a short time.In this paper,we construct three categories(eight kinds included)mushroom data sets using network crawler technology,and on this basis,establish an AI classification model.Finally,the trained classification model is transplanted into the mobile App to realize the real-time classification and recognition of mushrooms.The main work of this paper is as follows:(1)Construct the mushroom data sets with three categories(eight kinds included): use the web crawler tool to crawl eight kinds of mushroom images in the most frequently used internet search engines,then establish a data set based on the result above.(2)Training the classification models base on the data set: According to the characteristics of the mushroom data set built by myself,using Keras/TensorFlow deep learning framework to construct my multi-layer convolution neural network and use transfer-learning method on other classical convolution neural networks,the final trained model achieves 99% accuracy on the training set,and 89% accuracy on the validation set.(3)Transplant the established model into the iOS platform App: convert the trained model from.H5 file to.mlmodel file,build the iOS App's interactive interface under Xcode,and deploy.mlmodel model file into the App.(4)App testing and function improvement: Randomly select pictures from the Internet and data sets to test the recognition accuracy of the completed App,the App can achieve second-recognition and has a high recognition accuracy.Develop the assistant function based on the actual use demand for the current version of the App.