CN-119555625-B - Terahertz wireless sensing-based doped edible oil detection method and system

CN119555625BCN 119555625 BCN119555625 BCN 119555625BCN-119555625-B

Abstract

The invention provides a method and a system for detecting doped edible oil based on terahertz wireless sensing, wherein the method comprises the steps of obtaining edible oil to be detected, transmitting terahertz signals to the edible oil to be detected by using terahertz equipment, recording reflected signals, inputting the reflected signals of the edible oil to be detected into a pre-trained doped edible oil detection model, extracting global features of the reflected signals by an edible oil type feature extractor in a type identification module, analyzing the global features by the type detector to obtain edible oil types, extracting proportion features of the reflected signals by a proportion feature selector in a proportion quantization module, and comparing proportion feature analysis by a proportion quantizer to obtain edible oil adulteration proportion. The invention can realize high-sensitivity edible oil type doped identification and proportion quantification and has higher universality.

Inventors

MA HUADONG
ZHOU ANFU
SONG DENGHUI

Assignees

北京邮电大学

Dates

Publication Date: 20260505
Application Date: 20240905

Claims (7)

1. The method for detecting the doped edible oil based on terahertz wireless sensing is characterized by comprising the following steps of: Acquiring edible oil to be detected, transmitting terahertz signals to the edible oil to be detected by using terahertz equipment, and recording reflection signals of the edible oil to be detected; inputting the reflected signals of the edible oil to be detected into a pre-trained doped edible oil detection model to obtain the type and the adulteration proportion of the edible oil to be detected; The pre-training method of the doped edible oil detection model comprises the following steps of: Transmitting terahertz signals to an edible oil sample by using terahertz equipment, and recording sample reflection signals, wherein the edible oil sample comprises a pure edible oil sample and an adulterated edible oil sample; acquiring an initial detection model, wherein the initial detection model comprises a type identification module and a proportion quantization module, the type identification module comprises an edible oil category feature extractor and a category detector, and the proportion quantization module comprises a proportion feature selector and a proportion quantizer; Inputting a reflected signal of a pure edible oil sample into the edible oil characteristic extractor, extracting to obtain all pure edible oil sample characteristics, clustering all the pure edible oil sample characteristics to obtain a clustering center of each pure edible oil characteristic, and taking the clustering center as an anchor point of each pure edible oil characteristic; inputting the reflected signals of the adulterated edible oil samples into the edible oil characteristic extractor, extracting to obtain the characteristics of each adulterated edible oil sample, and carrying out characteristic similarity on the characteristics of each adulterated edible oil sample and the midpoint of the connecting line of the characteristic anchor points of the pure edible oil contained in each adulterated edible oil sample according to the pre-marked information so as to obtain the relation between the adulterated edible oil sample and the pure edible oil contained in each adulterated edible oil sample; the method comprises the steps that a double teacher network is adopted to carry out knowledge distillation training on edible oil characteristic extractors, wherein a first teacher network is a pure edible oil characteristic extractor which is obtained by training a pure edible oil sample in advance, a second teacher network is an adulterated edible oil characteristic extractor which is obtained by training an adulterated edible oil sample in advance, reflection signals of the pure edible oil and the adulterated edible oil sample are respectively input into the double teacher network and the edible oil characteristic extractor, category characteristic similarity is carried out, and the edible oil category characteristic extractors can extract pure edible oil characteristics and adulterated edible oil characteristics; The pre-training of the proportion quantization module comprises the steps of dividing edible oil into expensive edible oil and low-price edible oil in advance according to market value, inputting a reflected signal of an adulterated edible oil sample containing the same kind of expensive edible oil into a proportion characteristic selector to be trained, and selecting common proportion characteristics which are not interfered by the kind of the adulterated low-price edible oil and are only related to the adulterated proportion of the same kind of expensive edible oil through a multi-source domain self-adaptive method; and training the initial detection model by adopting the training set until the preset performance requirement is met, so as to obtain the edible oil doped detection model.
2. The method for detecting the edible oil doped based on terahertz wireless sensing according to claim 1, wherein after the terahertz device is used to transmit the terahertz signal to the edible oil to be detected and record the reflected signal of the edible oil to be detected, further comprises: and carrying out Fourier transform processing on the edible oil reflection signal to be detected, and converting the edible oil reflection signal to be detected from a time domain signal to a frequency domain signal.
3. The method for detecting the edible oil doped based on terahertz wireless sensing according to claim 1, wherein the edible oil classification feature extractor extracts global features of the sample reflected signals, and the class detector analyzes the global features to obtain the edible oil type, comprising: The edible oil other feature extractor extracts global features of the sample reflected signal based on a self-attention mechanism; And inputting the global characteristic into the category detector, and inputting the edible oil type by a full connection layer in the category detector.
4. The terahertz wireless sensing-based doped edible oil detection method according to claim 1, wherein the edible oil classification feature extractor is subjected to knowledge distillation training by adopting a double-teacher network, and the method comprises the following steps: Inputting the reflected signals of the pure edible oil sample into an initial class feature extractor, and training to obtain the pure edible oil class feature extractor; inputting the reflected signals of the adulterated edible oil sample into an initial class feature extractor, and training to obtain the adulterated edible oil class feature extractor; taking the pure edible oil unique feature extractor and the adulterated edible oil unique feature extractor as two teacher networks; The method comprises the steps of inputting reflected signals of a pure edible oil sample into a pure edible oil characteristic extractor and an edible oil characteristic extractor of a pure edible oil teacher network respectively, enabling the edible oil characteristic extractor to extract pure edible oil characteristics through category characteristic similarity, inputting reflected signals of an adulterated edible oil sample into an adulterated edible oil characteristic extractor and an edible oil characteristic extractor of an adulterated edible oil teacher network respectively, and enabling the initial edible oil characteristic extractor to extract adulterated edible oil characteristics through category characteristic similarity.
5. The method for detecting the doped edible oil based on terahertz wireless sensing according to claim 1, wherein the common characteristics of different doping ratios of the same kind of expensive edible oil are selected by using a selector, further comprising: The selector obtains deep features and shallow features based on learning and principal component analysis modes respectively to realize selection of proportional features.
6. A terahertz wireless sensing based doped edible oil detection system, characterized in that the system, when executed, implements the steps of the terahertz wireless sensing based doped edible oil detection method according to any one of claims 1 to 5, the system comprising: The data processing module is used for transmitting terahertz signals to the edible oil to be detected by using terahertz equipment and recording reflection signals of the edible oil to be detected; The detection module is used for inputting the reflected signals of the edible oil to be detected into a pre-trained doped edible oil detection model to obtain the type and the adulteration proportion of the edible oil to be detected.
7. A computer readable storage medium having stored thereon a computer program/instruction which when executed by a processor performs the steps of the method according to any of claims 1 to 5.

Description

Terahertz wireless sensing-based doped edible oil detection method and system Technical Field The invention relates to the technical field of wireless sensing, in particular to a terahertz wireless sensing-based doped edible oil detection method and system. Background Edible oil is an important source of fat intake in humans and is critical to maintaining nutritional balance and avoiding non-infectious diseases such as obesity, malnutrition, etc. The adulterated edible oil not only fool the consumer, but may also be a long-term hazard to human health, for example, ingestion of certain trans fatty acids may increase the risk of coronary heart disease. Therefore, the detection of the adulterated edible oil has important practical significance and necessity for daily life. Because the major components of edible oils are relatively similar, it is difficult to directly identify the adulterated edible oil by sensory methods such as color, odor, consistency, etc. Therefore, a reliable technology for detecting the adulterated edible oil is to detect the physical molecular property of the edible oil, wherein the prior detection technical scheme comprises the steps of separating and identifying the molecular types in an edible oil sample by utilizing the differential interaction between a compound and a stationary phase and a mobile phase through a first high performance liquid chromatography (High Performance Liquid Chromatography, HPLC) and a gas chromatography (Gas Chromatography, GC), analyzing the characteristic absorption modes of different molecular groups and chemical bonds in different spectral ranges through a second near infrared (NEAR INFRARED, NIR) spectrum to infer the molecular types, distinguishing the molecular types by utilizing the slight difference of molecular mass through a third mass spectrometry, distinguishing the molecular types by utilizing a nuclear magnetic resonance (Nuclear Magnetic Resonance Imaging, NMR) instrument, an electronic tongue and a fluorescent marker, and evaluating whether the edible oil is adulterated or not. However, these prior art solutions above generally require specialized equipment and expertise of trained professionals to operate, which makes them unusable for everyday use in everyday environments. In recent years, wireless signal sensing technologies, such as radio frequency identification (Radio Frequency Identification, RFID), wiFi, ultra Wide Band (UWB), millimeter wave (mmWave), and the like, have been rapidly developed under the promotion of advantages such as speed, non-invasiveness, non-destructiveness, and the like. Accordingly, various non-contact sensing technologies using wireless signals have emerged. For example, attempts have been made to classify edible oils using signals from ultra-bandwidth and millimeter waves. However, due to inherent limitations in their wavelength, these signals may not carry a sufficient amount of identifying information about the different oil types and are therefore less suitable for detection of doped edible oils. Disclosure of Invention In view of this, the embodiment of the invention provides a method and a system for detecting doped edible oil based on terahertz wireless sensing, so as to eliminate or improve one or more defects existing in the prior art. The invention provides a terahertz wireless sensing-based edible oil doped detection method, which comprises the following steps of: Acquiring edible oil to be detected, transmitting terahertz signals to the edible oil to be detected by using terahertz equipment, and recording reflection signals of the edible oil to be detected; inputting the reflected signals of the edible oil to be detected into a pre-trained doped edible oil detection model to obtain the type and the adulteration proportion of the edible oil to be detected; The pre-training method of the doped edible oil detection model comprises the steps of transmitting terahertz signals to an edible oil sample by using terahertz equipment, recording sample reflection signals, marking the sample reflection signals, constructing a training set by using the edible oil sample including a pure edible oil sample and a doped edible oil sample, acquiring an initial detection model, wherein the initial detection model comprises a type identification module and a proportion quantization module, the type identification module comprises an edible oil characteristic extractor and a type detector, the proportion quantization module comprises a proportion characteristic selector and a proportion quantizer, inputting the sample reflection signals into the initial detection module, extracting global characteristics of the sample reflection signals in the type identification module, analyzing the global characteristics by the type detector to obtain an edible oil type, extracting proportion characteristics of the sample reflection signals in the proportion quantization module, analyzing the proportion characteristics by the propo