Search

CN-122003624-A - Computer-implemented method, system and computer program for detecting the presence of metal in an object

CN122003624ACN 122003624 ACN122003624 ACN 122003624ACN-122003624-A

Abstract

The invention relates to a computer-implemented method for detecting the presence of metal in an object (100), the object (100) being transported along a transport path through a metal detector unit (10), preferably through a belt conveyor (101), the method comprising providing detection data to an electronic data processing device (20), the detection data being created by the metal detector unit (10), the metal detector unit (10) comprising a source unit (11) for an electromagnetic field and a receiver unit (12) being electromagnetically coupled to the source unit, wherein the source unit (11), the receiver unit (12) and the transport path are arranged relative to each other such that an electrical signal depending on the metal content in the object (100) can be output, the metal detector unit (10) further being operative to derive the detection data from the electrical signal, processing the provided detection data by a metal detection software module executed by the electronic data processing device (20,331,421), the metal detection software module comprising a first machine learning module (22) receiving the detection data as a first machine learning module and being able to recognize that the metal is present in the first machine learning module (21) and being able to recognize the presence of the metal in the first machine learning module (21) based on the first machine learning module, the invention also relates to a system and a computer program for detecting the presence of metal in an object (100).

Inventors

  • TAO YANG

Assignees

  • 梅特勒-托利多安全线有限公司

Dates

Publication Date
20260508
Application Date
20241122
Priority Date
20231127

Claims (15)

  1. 1. A computer-implemented method for detecting the presence of metal in an object (100), the object (100) being transported along a transport path through a metal detector unit (10), preferably a belt conveyor (101), the method comprising: Providing detection data to an electronic data processing device (20), the detection data being created by a metal detector unit (10), the metal detector unit (10) comprising a source unit (11) for an electromagnetic field and a receiver unit (12) electromagnetically coupled to the source unit, wherein the source unit (11), the receiver unit (12) and the conveying path are arranged relative to each other such that an electrical signal depending on the metal content within the object (100) can be output, the metal detector unit (10) further being operative to derive the detection data from the electrical signal; The provided detection data is processed by a metal detection software module executed by an electronic data processing device (20,331,421), the metal detection software module comprising a first machine learning module (22), It is characterized in that the method comprises the steps of, The first machine learning module receives the detection data as input and has been trained to automatically identify the presence of metal in the object (10) based on the detection data, the first machine learning module (21) being further configured to output identification data indicative of the presence of metal identified by the first machine learning module (22).
  2. 2. The method of claim 1, wherein the training comprises creating the detection data at a remote location (310, 410) and transmitting the created detection data to a host location (330,420) where the created detection data is used as training data for machine learning, thereby training a first machine learning module (22), the first machine learning module (22) being deployed to an electronic data processing device (20,331,421).
  3. 3. The method according to claim 1 or 2, wherein the first machine learning module (22) comprises a neural network, preferably a deep neural network, trained for automatically identifying the presence of metal in an object (100) based on the detection data by using marker training data.
  4. 4. A method according to claim 3, wherein the marker training data comprises metal-free training data associated with at least one metal-free training object (100) and/or metal training data associated with at least one metal-containing training object, wherein the metal-free training data comprises detection data derived by a metal detector unit (10) when the at least one metal-free training object (100) is conveyed through the metal detector unit (10), and/or wherein the metal training data comprises synthetic data created by a training data generation software module.
  5. 5. The method of claim 3 or 4, wherein the neural network is further configured to generate an activation mask (600) for the detection data, the activation mask indicating an activation level at each instance of the detection data, thereby predicting a location of metal in the object (100).
  6. 6. The method of any of the preceding claims, wherein the first machine learning module (22) is further trained to identify an object class of an object (100) from the detection data and to output object class data indicative of the identified object class.
  7. 7. The method according to any of the preceding claims, wherein the metal detection software module comprises a filter function for applying interference or noise filtering to the provided detection data, and/or the first machine learning module (22) has been trained by using detection data contaminated with interference or noise effects.
  8. 8. The method of any of the preceding claims, further comprising: Collecting, by the sensor unit (30), operational status data indicative of an operational status of the metal detector unit (10); Providing the operational state data to an operational state software module executed by an electronic data processing device (20,331,421), the operational state software module comprising a second machine learning module (23) that has been trained by machine learning to be able to automatically identify anomalies in the operational state data, wherein the second machine learning module (23) preferably comprises a neural network, more preferably a deep neural network, for automatic identification of the anomalies, the operational state software module being configured to be able to receive the operational state data as input and to output anomaly data indicative of the presence of anomalies identified in the operational state data by the second machine learning module (23).
  9. 9. The method according to any of the preceding claims, wherein the connection between the metal detector unit (10) and/or the electronic data processing device (20,331,421) and/or the sensor unit (10) and/or the host location (330,420) is realized via a network, preferably via a wireless network, so that data, in particular the detection data and/or the training data and/or the identification data and/or the operating state data, can be exchanged.
  10. 10. A method according to any one of the preceding claims, wherein the source unit (11) comprises a transmitter coil to which an alternating current is applied to generate a time dependent electromagnetic field, the receiver unit (12) comprises a balanced coil system, the electrical signal is a differential voltage output of the balanced coil system as a function of time, and the detection data preferably comprises a time sequence of data points, the data points being real and imaginary parts of a phasor expression of the function of the differential voltage signal at different points in time.
  11. 11. The method according to any one of the preceding claims, wherein a time-dependent input signal is applied to a source unit (10) generating a time-dependent electromagnetic field, and wherein the first machine learning module (22) is further trained to be able to determine a reference phase of the electrical signal from the input signal based on the detection data.
  12. 12. The method of any one of the preceding claims, wherein the first machine learning module (22) comprises a probability density function of metal-free detection data, the probability density function having been generated by machine learning using metal-free training data associated with a metal-free object, the first machine learning module (22) being further operable to calculate a probability value of the probability density function for the received detection data, and wherein the identification data comprises data indicative of the presence of metal in the object (100) if the probability value is above or below a predetermined threshold.
  13. 13. The method of claim 12, wherein generating the probability density function of the first machine learning module (22) by machine learning comprises the steps of: Providing a probability density model function of the detection data parameterized by the set of model parameters; Conveying a plurality of metal-free training objects (100) through a metal detector unit (10) creating metal-free training data comprising detection data created for the metal-free training objects (100); Determining values of model parameters such that probability density functions of probability density model functions defined as values of the model parameters are consistent with the metal-free training data and maximize likelihood; The threshold is defined given a probability density function.
  14. 14. A system (1) for detecting the presence of metal in an object (100), the system (1) comprising: A metal detector unit (10) comprising a source unit (11) for an electromagnetic field and a receiver unit (12) electromagnetically coupled to the source unit; a conveying device (101), preferably a belt conveyor, for conveying the objects (100) along a conveying path through the metal detector unit (10); Wherein the source unit (11), the receiver unit (12) and the conveying path are arranged relative to each other such that an electrical signal depending on the metal content within the object (100) can be output, the metal detector unit (10) being further operative to derive detection data from the electrical signal; An electronic data processing device (20,331,421) configured to receive the detection data as input and to process the provided detection data by executing a metal detection software module, the metal detection software module comprising a first machine learning module (22), It is characterized in that the method comprises the steps of, The first machine learning module receives the detection data as input and has been trained to automatically identify the presence of metal in an object (100) based on the detection data, the first machine learning module (22) being further configured to output identification data indicative of the presence of metal identified by the first machine learning module (22).
  15. 15. A computer program for detecting the presence of metal in an object (100), the object (100) being conveyed along a conveying path through a metal detector unit (10), preferably a belt conveyor (101), the metal detector unit (10) comprising a source unit (11) for an electromagnetic field and a receiver unit (12) being electromagnetically coupled to the source unit, wherein the source unit (11), the receiver unit (12) and the conveying path are arranged relative to each other such that an electrical signal depending on the metal content within the object (100) can be output, the metal detector unit (10) further being operative to derive detection data from the electrical signal, the computer program comprising instructions which, when executed by a computer, cause the computer to perform the steps of: receiving the detection data as input; the provided detection data is processed by a metal detection software module comprising a first machine learning module (22), It is characterized in that the method comprises the steps of, The first machine learning module receives the detection data as input and has been trained to automatically identify the presence of metal in the object (100) based on the detection data, the first machine learning module (22) being further configured to output identification data indicative of the presence of metal identified by the first machine learning module (22).

Description

Computer-implemented method, system and computer program for detecting the presence of metal in an object Technical Field The present invention relates to a computer implemented method, system and computer program for detecting the presence of metal in an object. Background Detecting the presence of metal in an object is of great importance in various industries, for example in the food and pharmaceutical industries. In these industries, metal detectors are used to detect metal contaminants in edible goods and other products, thereby improving product safety. One example of a system known in the art for detecting the presence of metal in an object includes a metal detector unit having a source unit for an electromagnetic field and a receiver unit electromagnetically coupled to the source unit, and a conveying device (e.g., a belt conveyor or a gravity-fed conveying device) for conveying the object along a conveying path through the metal detector unit. The source unit, the receiver unit and the conveying path are arranged relative to each other such that an electrical signal dependent on the metal content in the object can be output, and the metal detector unit is further operative to derive detection data from the electrical signal for further data processing. Such metal detection systems are disclosed, for example, in US 8,841,903 B2 and US 10,989,829 B2. The detection data may be provided to an electronic data processing device for further data processing. The electronic data processing device executes a discrimination algorithm, for example a distance-based discrimination algorithm (see e.g. EP 2 625 A1 or EP 3 726 256 A1), which is configured to be able to detect the presence of metal in the object based on said detection data. In WO 2022/071700 A2 and KR 102 475 911 B1, the detection data is input into an image generator to form a two-dimensional image (photo). The image is provided as an input to a deep learning machine that has been trained to be able to detect the presence of metal in an object based on the image. Metal detection using machine learning is known in the field of metal detection using unmanned aerial vehicles (see "Optimization: Drone-Operated Metal Detection based on Machine Learning and PID Controller", International Journal of Precision Engineering and Manufacturing, Korean Society for Precision Engineering, Springer, Vol. 23, No. 5, 503-515 (2022)） of j Minho et al and the field of magnetoresistive sensors (see also "Neural Network for Metal Detection Based on Magnetic Impedance Sensor", Sensors, Vol. 21, No. 13, 4456 (2021)） of h Sungjae et al US 2018/243800 A1 relates to a material sorting system using a vision system, whereas EP 4 089 495 A1 relates to an analysis device and method for monitoring the operating state of an industrial system. Conventional metal detectors are typically deployed independently, however in the context of internet of things and deep learning, the metal detectors will communicate with each other, even with a central server, which enables more powerful functions and capabilities. CN 115 327 647A discloses a portable wireless metal detector that communicates with a cloud or remote server over a network. The detection of smaller and smaller metal contaminants in objects is a common theme of the metal detection industry. However, a trade-off is typically required between the sensitivity of the detector unit (i.e. its discrimination accuracy) and false positives due to product effects (i.e. the inspected object may produce a more pronounced signal than the signal of small metal contaminants). It is therefore an object of the present invention to provide a method, system and computer program for detecting the presence of metal in an object, which provides high performance, in particular high discrimination accuracy. Disclosure of Invention According to a first aspect of the present invention there is provided a computer-implemented method for detecting the presence of metal in an object conveyed along a conveying path through a metal detector unit, preferably a belt conveyor, the method comprising providing detection data to an electronic data processing apparatus, the detection data being created by the metal detector unit, the metal detector unit comprising a source unit for an electromagnetic field and a receiver unit electromagnetically coupled to the source unit, wherein the source unit, the receiver unit and the conveying path are arranged relative to each other such that an electrical signal dependent on the metal content in the object can be output, the metal detector unit further being operative to derive the detection data from the electrical signal, processing the provided detection data by a metal detection software module executed by the electronic data processing apparatus, the metal detection software module comprising a first machine learning module which receives the detection data as input and has been trained to be able to automatically id