Chinese mitten crab freshness damage-free detection method based on semi-supervised identification projection

Abstract

The invention discloses a Chinese mitten crab freshness damage-free detection method based on semi-supervised identification projection. The detection method includes: pre-processing a sample and pre-heating an electronic nose apparatus; sampling the sample by deeply putting a sample needle of the electronic nose apparatus into a beaker; collecting smell data collected by a sensor array in the electronic nose apparatus, pre-processing the data, and performing character recognition process to the pre-processed sensor original data. In the method, low-dimension feature vectors extracted through the semi-supervised identification projection algorithm serves as an input mode vector of a self-organizing feature mapping neural network, thereby forming a prediction model for level of freshness of the Chinese mitten crabs. The method is used for damage-free detection of the sample and is greatly improved in detection precision, and can achieve quick and damage-free detection on the quality of the Chinese mitten crabs.

Description

technical field [0001] The invention provides a hairy crab freshness nondestructive detection method based on semi-supervised discrimination projection, and relates to the technical field of detection methods. Background technique [0002] Hairy crab is a famous delicacy because of its delicious taste and rich nutrition. It is widely distributed not only in my country, but also in Europe and North America. But for the hairy crabs that are not fresh, due to the decomposition of their own enzymes and the growth of various bacteria, the meat of the hairy crabs will deteriorate, causing sour and smelly fermentation, and slowly volatilize nitrogen-containing products, amines, ammonia, and alcohols And sulfur-containing products and other gases with corrupt characteristics, sometimes even produce toxic substances such as histamine, which cannot be destroyed under high-temperature cooking. If food poisoning is caused by eating, the consequences will be unimaginable. [0003] At pr...

AI Technical Summary

Problems solved by technology

Linear dimensionality reduction algorithms such as PCA and LDA are characterized by linearly mapping high-dimensional data to a low-dimensional subspace in the form of matrix operations. The advantage is that the learning process and dimensionality reduction process have low computational complexity and fast operation speed. The disadvantages are Dimensionality reduction is not effective for data that is essentially a nonlinear distribution

The clustering effect obtained by clustering-based Cluster-then-Label is not obvious when the sample size is large

However, nonlinear dimensionality reduction algorithms such as KPCA do not consider the category information of samples.

In addition, most of these algorithms analyze the electronic nose response data in the Euclidean space, and cannot fully mine the potential information contained in high-dimensional data.

Moreover, the difference in the odor information volatilized during the storage of live hairy crabs is small. In addition, the sensor array generally has problems such as baseline drift and poor repeatability. As a result, these traditional feature extraction algorithms cannot accurately judge the freshness of hairy crabs.

Images