CN-121995016-A - Cereal detection device and detection method

CN121995016ACN 121995016 ACN121995016 ACN 121995016ACN-121995016-A

Abstract

The invention relates to the technical field of grain detection, in particular to a grain detection device and a grain detection method, comprising the steps of acquiring original detection data and carrying out association processing to obtain associated image data and associated spectrum data; the method comprises the steps of carrying out particle segmentation and invalid particle screening on associated image data to determine an effective particle set, extracting image features and spectrum features of each effective particle based on the associated image data and associated spectrum data to form initial features, carrying out initial risk candidate screening to obtain a candidate risk particle set and initial risk information, and constructing a risk distribution result and a mildew detection result. According to the scheme, the whole batch of average signals are refined to particle-level risk identification, local abnormal information is reserved under the condition of not depending on large-scale crushing and sample mixing, the whole batch screening accuracy is improved, the risk that the abnormality is covered under the conditions of local aggregation and in-batch uneven distribution is reduced, and the detection efficiency and the result stability are both considered.

Inventors

LI XIANGDONG
XU DAHAI
WANG ZHEN

Assignees

山东才聚电子科技有限公司

Dates

Publication Date: 20260508
Application Date: 20260410

Claims (10)

1. The grain detection device comprises a pneumatic conveying device, an impurity detection device, a moisture detection device, a volume weight detection device and a mildew particle detection device, wherein the volume weight detection device is arranged on the lower side of the moisture detection device, the pneumatic conveying device respectively conveys grains to the impurity detection device, the moisture detection device and the mildew particle detection device, and the grain detection device is characterized by comprising a grain conveying mechanism and a data acquisition component, the grain conveying mechanism is conveyed by the pneumatic conveying device, the grains are flatly conveyed on the grain conveying mechanism, the data acquisition component is arranged on the upper side of the grain conveying mechanism, or the data acquisition component is arranged on the two sides of falling grains after being conveyed by the grain conveying mechanism, and the data acquisition component acquires image data and spectrum data.
2. The grain detecting method of the grain detecting apparatus according to claim 1, wherein the impurity content detection, the moisture detection, the bulk density detection and the mildew particle detection are performed on the grain by using the impurity detecting apparatus, the moisture detecting apparatus, the bulk density detection and the mildew particle detection, respectively, wherein the mildew particle detection method is as follows: the data acquisition component acquires original detection data, and carries out association processing on the original detection data to obtain association detection data, wherein the association detection data comprises association image data and association spectrum data; Performing particle segmentation and ineffective particle screening on the associated image data to determine an effective particle set; Performing image feature extraction and spectrum feature extraction on each effective particle in the effective particle set based on the associated image data and the associated spectrum data, and establishing a corresponding relation to form initial features of each effective particle; Screening initial risk candidates based on the initial characteristics to obtain a candidate risk particle set, and determining initial risk information corresponding to each candidate risk particle; And constructing a risk distribution result for the candidate risk particle set based on the initial risk information, and generating a mildew detection result according to the risk distribution result.
3. The method for detecting grains according to claim 2, wherein the specific steps of obtaining the original detection data and performing the association processing on the original detection data to obtain the association detection data include: Acquiring original detection data obtained by continuously acquiring single-layer conveyed cereal batch samples in a target area by a data acquisition component, wherein the original detection data comprises original image data and original spectrum data; Performing unified time reference processing on the original image data and the original spectrum data to obtain an image data set and a spectrum data set; Respectively carrying out buffer management and sequential arrangement on the image dataset and the spectrum dataset to form an image buffer queue and a spectrum buffer queue; and matching and merging the image cache queue and the spectrum cache queue to obtain the associated detection data.
4. The grain detection method of claim 2, wherein the specific steps of performing grain segmentation and ineffective grain screening on the associated image data to determine an effective set of grains comprise: preprocessing and foreground extraction are carried out on the associated image data to obtain a particle communication area; determining region categories based on region features of the particle connected regions, and screening out the region categories to obtain normal candidate regions and adhesion regions; The normal candidate region is reserved as a first effective particle, and the adhesion region is split and subjected to effectiveness judgment to obtain a second effective particle; and performing frame-crossing de-duplication and abnormal elimination on the first effective particles and the second effective particles to obtain an effective particle set.
5. The method for detecting grains according to claim 2, wherein the specific step of forming the initial characteristic of each effective grain comprises: establishing a particle characteristic extraction queue based on the effective particle set and the associated detection data, and calling a mapping table from a pre-established image column coordinate to a spectrum sampling column; searching spectrum sampling columns corresponding to each effective particle based on the mapping table and the particle characteristic extraction queue to form a spectrum sampling column set; Judging whether each effective particle and the spectrum sampling column set establish a corresponding relation, executing the complement matching and abnormal rejection on the effective particles which do not establish the corresponding relation, extracting image features and spectrum features of the effective particles which establish the corresponding relation, merging extraction results, and forming initial features of each effective particle.
6. The method for detecting grains according to claim 2, wherein the specific steps of screening the candidate risk candidates based on the initial characteristics to obtain a candidate risk particle set, and determining initial risk information corresponding to each candidate risk particle include: based on the initial characteristics of each effective particle, normalizing the image characteristics and the spectrum characteristics of each effective particle in the same batch, determining the abnormal characteristic quantity and the spectrum image joint abnormal quantity, and determining an initial risk score according to the spectrum image joint abnormal quantity; Performing initial risk candidate screening on each effective particle according to the initial risk scores and the abnormal feature quantity to obtain a candidate risk particle set; And generating initial risk information according to the candidate risk particle set.
7. The grain detection method according to claim 2, wherein the specific steps of constructing a risk distribution result for the candidate risk particle set based on the initial risk information, and generating a mildew detection result according to the risk distribution result include: constructing a risk statistical grid based on the candidate risk particle set, the initial risk information and the effective particle set, calculating grid risk density, merging adjacent risk statistical grids, and forming a local risk region; Counting risk distribution characteristics based on the local risk areas, and carrying out threshold judgment according to the risk distribution characteristics, and constructing a risk distribution result after the judgment is passed; Generating a mildew detection result according to the risk distribution result.
8. The method for detecting cereal according to claim 7, further comprising: Constructing a non-toxin disturbance characterization for the particles in the candidate risk particle set; extracting toxin-related abnormal characteristics in the candidate risk particle sets, and carrying out decoupling treatment by combining the non-toxin disturbance characterization to obtain toxin risk representations corresponding to each candidate risk particle; and correcting the risk distribution result according to the toxin risk representation to form a corrected distribution result, and generating a corrected detection result based on the corrected distribution result.
9. The grain detection method of claim 8, wherein the specific step of constructing a non-toxin perturbation signature for the particles in the candidate risk particle set comprises: Carrying out regional normalization and extracting non-toxin disturbance sub-representations based on initial features corresponding to the candidate risk particle sets, wherein the non-toxin disturbance sub-representations comprise water content disturbance, particle variety disturbance, particle size disturbance, surface damage disturbance and background illumination disturbance; Merging the non-toxin disturbance rotor representations according to weights to obtain non-toxin disturbance representations corresponding to each candidate risk particle; calibrating the disturbance state of each candidate risk particle according to the non-toxin disturbance characterization; The specific steps of extracting the toxin-related abnormal characteristics in the candidate risk particle set and carrying out decoupling treatment by combining the non-toxin disturbance characterization to obtain toxin risk representations corresponding to each candidate risk particle comprise: Extracting toxin-related abnormal characteristics in the candidate risk particle sets, and weighting the toxin-related abnormal characteristics to obtain an original toxin abnormal value; According to the non-toxin disturbance characterization, carrying out disturbance state division on candidate risk particles, endowing a non-toxin deduction coefficient based on the disturbance state, and carrying out decoupling treatment on an original toxin abnormal value and the non-toxin disturbance characterization to obtain a toxin risk representation; According to the toxin risk representation, the particle states of candidate risk particles are formed, and the risk distribution result is corrected based on the particle states, wherein the particle states comprise toxin dominant particles, toxin particles to be rechecked and non-toxin dominant particles.
10. The grain detection method of claim 9, wherein the specific steps of correcting the risk distribution result according to the toxin risk representation, forming a corrected distribution result, and generating a corrected detection result based on the corrected distribution result include; recalculating grid risk densities of toxin leading particles and toxin particles to be rechecked in each risk statistical grid based on the risk statistical grids, wherein the toxin leading particles and the toxin particles to be rechecked form corrected risk particles; Merging the risk statistical grids adjacent longitudinally or transversely to form a corrected risk area; And constructing a correction distribution result according to the correction risk particles and the correction risk areas, and generating a correction detection result according to the correction distribution result.

Description

Cereal detection device and detection method Technical Field The invention relates to the field of grain detection, in particular to a grain detection device and a grain detection method. Background In the warehouse entry process of a grain depot, the warehouse entry grains need to be pre-screened, the grain mildew condition is detected, the grain safety is ensured, and as mildew pollution in the whole grains generally presents the characteristics of low concentration, local aggregation and uneven distribution in the whole grains, and different grains have obvious fluctuation in the aspects of water content, surface integrity, variety difference, mildew degree, site illumination condition and the like, the existing detection scheme is difficult to simultaneously consider the whole screening efficiency, low-concentration local pollution recognition capability and detection result stability under the condition of not carrying out large-scale crushing and sample mixing, and the condition that the whole average signal of the whole grains is normal but the local high-risk grains are covered by the whole signal is easy to appear. At present, the prior art generally adopts the modes of chromatographic mass spectrometry, near infrared spectrum, hyperspectral imaging or Raman and the like to carry out laboratory detection or on-site rapid screening on grain samples, and in an artificial intelligent processing path, the collected spectrum data, image data or spectrum and image fusion data are generally subjected to pretreatment, feature extraction, feature selection and model training, and then a classification model or regression model is utilized to judge whether mildew exceeds standard or estimate mycotoxin content so as to realize rapid identification and risk judgment on the grain samples. However, in the prior art, in practical application, single-sample average spectrum, single-view-field average characteristics or homogenized integral signals are still used as main analysis objects, and although the mode is convenient for model training and quick discrimination, the detection result mainly reflects integral average differences, and it is difficult to accurately embody localization, sparsification and uneven spatial distribution characteristics of toxin pollution in the whole grains. For example, when the overall signal of the whole batch of cereal grains appears substantially normal, but with small amounts of high risk particles entrained therein, these local anomaly signals that truly determine the safety level of the whole batch are easily masked by the overall average result, resulting in insufficient recognition of local high risk contaminations, with risk of missed detection. Disclosure of Invention The invention aims to solve the problems that mildew pollution in whole grains often presents a low concentration, local aggregation and uneven distribution in the grains, and the prior art is mainly judged by using an overall average signal, so that a small amount of local high-risk grains are easily covered and the detection omission risk exists. In order to achieve the above purpose, in one aspect, the invention provides a grain detection device, which comprises a pneumatic conveying device, an impurity detection device, a moisture detection device, a volume weight detection device and a mildew particle detection device, wherein the volume weight detection device is arranged at the lower side of the moisture detection device, the pneumatic conveying device respectively conveys grains to the impurity detection device, the moisture detection device and the mildew particle detection device, and the grain detection device is characterized by comprising a grain conveying mechanism and a data acquisition component, the pneumatic conveying device conveys the grains to the grain conveying mechanism, the grains are flatly conveyed on the grain conveying mechanism, the data acquisition component is arranged at the upper side of the grain conveying mechanism, or the data acquisition component is arranged at the two sides of falling grains after being conveyed by the grain conveying mechanism, and the data acquisition component acquires image data and spectrum data. On the other hand, the invention provides a grain detection method of the grain detection device, which utilizes an impurity detection device, a moisture detection device, a volume weight detection device and a mildew particle detection device to respectively detect impurity content, moisture, volume weight and mildew particle, wherein the mildew particle detection method comprises the following steps: the data acquisition component acquires original detection data, and carries out association processing on the original detection data to obtain association detection data, wherein the association detection data comprises association image data and association spectrum data; Performing particle segmentation and ineffective particle screening on the associate