Deep Learning and Blockchain Based Traceability of Boletus

Abstract

Boletus have become popular in recent years. However, as a kind of food, its edible safety has been widely concerned. Therefore, it is necessary to trace its origin. In this paper, a total of 1195 samples of 8 kinds of common boletus in southwest China were collected. Its Infrared Spectroscopy were obtained by FTIR spectrophotometer and converted into 2DCOS spectral images. The dataset was pre-processed and sent into ResNet-20 model for training. The model achieved an accuracy rate approaching $100 \%$ with minimal loss. By building a blockchain on the FISCO BCOS platform, a traceability framework for the supply chain of Boletus based on a consortium blockchain and a smart contract was established. Based on the results of ResNet-20 model, an Android app was developed and deployed using Java in Android Studio. After detecting Boletus products, upload the information to the blockchain to generate a traceability ID. Users can input these traceability IDs in the app to access Boletus information. Additionally, a visualization platform was developed using Echarts and Python, which utilizes Web3.js to retrieve and display relevant data from the blockchain. This research can be utilized as a point of reference for tracing Boletus.