CN113624759A - Apple nondestructive testing method based on machine learning - Google Patents

Apple nondestructive testing method based on machine learning Download PDF

Info

Publication number
CN113624759A
CN113624759A CN202110906221.5A CN202110906221A CN113624759A CN 113624759 A CN113624759 A CN 113624759A CN 202110906221 A CN202110906221 A CN 202110906221A CN 113624759 A CN113624759 A CN 113624759A
Authority
CN
China
Prior art keywords
apple
model
data
apples
nondestructive
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202110906221.5A
Other languages
Chinese (zh)
Inventor
郝红娟
王九鑫
刘宇程
卢定泽
苏耀恒
杨宁
吴鑫
李文龙
王康华
杜雨蓉
杨彤彤
王明墺
张倩
陈琳
张芷叶
黄磊
张亚鑫
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Xian Polytechnic University
Original Assignee
Xian Polytechnic University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Xian Polytechnic University filed Critical Xian Polytechnic University
Priority to CN202110906221.5A priority Critical patent/CN113624759A/en
Publication of CN113624759A publication Critical patent/CN113624759A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/84—Systems specially adapted for particular applications
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
    • G01N29/04—Analysing solids
    • G01N29/043—Analysing solids in the interior, e.g. by shear waves
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
    • G01N29/04—Analysing solids
    • G01N29/12—Analysing solids by measuring frequency or resonance of acoustic waves
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06T5/70—
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/84—Systems specially adapted for particular applications
    • G01N2021/8466—Investigation of vegetal material, e.g. leaves, plants, fruits
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N2291/00—Indexing codes associated with group G01N29/00
    • G01N2291/02—Indexing codes associated with the analysed material
    • G01N2291/023—Solids
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N2291/00—Indexing codes associated with group G01N29/00
    • G01N2291/02—Indexing codes associated with the analysed material
    • G01N2291/028—Material parameters
    • G01N2291/0289—Internal structure, e.g. defects, grain size, texture
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10004—Still image; Photographic image
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20024—Filtering details
    • G06T2207/20032—Median filtering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30181—Earth observation
    • G06T2207/30188—Vegetation; Agriculture

Abstract

The invention discloses an apple nondestructive testing method based on machine learning, and relates to a method for intelligently testing and grading apples by utilizing image analysis. The method comprises the following steps: obtaining an apple multi-parameter training model; setting an apple quality screening interval; the method comprises the steps of obtaining an appearance picture of an apple to be detected, recognizing the appearance picture of the apple by using a training model after denoising, carrying out internal nondestructive testing on the selected required apple, carrying out nondestructive judgment on the inside of the apple by using the training model, carrying out internal resonance sound wave detection on the disease-free apple, judging the content of internal substances by using the training model, and finishing classification of all apples. The method disclosed by the invention can be used for accurately detecting a large number of apples quickly and efficiently, so that the detection accuracy and efficiency are improved, and the detection flow is optimized. Not only can reduce the workload of the fruit grower, but also can improve the income of the fruit grower. And fruits with different grades can be provided according to the requirements of consumers, and certain effect is achieved on improving the position of domestic fruits.

Description

Apple nondestructive testing method based on machine learning
Technical Field
The invention relates to a method for intelligently detecting and grading apples by utilizing image analysis.
Background
The development of agriculture is closely related to the development of comprehensive national strength and the stability of national conditions in China, and the prosperous agriculture is one of the main targets at the present stage. China is currently in a key period for the transition from traditional agriculture to modern agriculture. Although the fruit resources are very abundant, the fruit yield already occupies the first position of the world, the export quantity and the quality are lower, the export quantity of the current fruit only occupies 2 percent of the world, and most of the fruit is sold in China. Taking apple as an example, one important factor influencing the development of the apple is that the grading and screening of the fruits are inaccurate, and the detection of the fruits cannot reach the export standard, so that the selling price of the fruits is influenced. The phenomenon of degrading primary products to secondary products, or upgrading secondary products to primary products, often occurs causing significant economic losses and consumer dissatisfaction. Because most fruits are sold in the country, the fruit detection device at the present stage mainly uses the traditional apple classification method, namely, the traditional apple classification method is identified by manpower and immature machines, screening is carried out according to the size and the appearance of the apples, the error rate is high, the efficiency is low, the influence of subjective feeling is large, the device cannot meet the requirements of the production and the sale of the existing apples, the quality of the fruits is reduced to some extent, and the market competitiveness is difficult to obtain comprehensive guarantee. This highlights the need to develop better automated apple classification techniques.
The nondestructive detection technology achieves the purposes of food detection and classification by signal acquisition and control and processing and analyzing the acquired signals on the basis of not damaging fruits. At the present stage, the research on nondestructive testing of fruits mainly uses the principles of optics, electromagnetism, mechanics, signal processing and the like, and focuses on the nuclear magnetic resonance technology, the dielectric property technology, the vibro-acoustic technology, the near infrared spectroscopy analysis technology, the machine vision technology and the X-ray technology, so that a plurality of researchers construct a system for detecting fruits by using the method. However, the systems have the problems that accurate identification cannot be realized, a large amount of rapid detection cannot be realized, and intellectualization cannot be realized. With the rapid development of artificial intelligence technology and mechanical automation technology, a machine learning technology is gradually popularized, and the deep learning algorithm has the advantages of strong acuity and wide available range. The deep learning technology and the nondestructive testing technology are deeply fused, so that the accurate screening and grading of fruits becomes an important trend for future development.
Disclosure of Invention
The invention aims to solve the technical problem of providing a machine learning-based apple nondestructive testing method aiming at the defects of the prior art. Not only can judge the appearance quality of the apples, but also can judge the content of sugar in the apples, such as the like, so that the quality of the apples can be screened and graded. The device disclosed by the invention realizes intellectualization as a whole, solves the problem that the prior art cannot accurately identify, and quickly, efficiently and accurately screens apples in a grading manner. The technical scheme is as follows:
a machine learning-based apple nondestructive testing method, the method comprising:
s0, collecting appearance pictures and size data of the apples, internal ultrasonic nondestructive data and internal resonance sound wave data, and training the data to obtain a training model;
s1, setting an apple quality screening interval;
s2, conveying the apples to an appearance nondestructive detection area;
s3, obtaining an appearance picture of the apple to be detected, removing salt and pepper noise of the picture, identifying the appearance picture of the apple by using the trained model, and determining the required apple according to the identification result;
s4, conveying the needed apples screened in the S3 to an internal nondestructive detection area, conveying irrelevant apples to an alternative selection area, and carrying out screening after parameters are reset;
s5, obtaining internal ultrasonic nondestructive data of the to-be-detected apple in the internal nondestructive detection area, carrying out nondestructive judgment on the interior of the apple by using the trained model, determining that the interior of the apple is free of diseases, if the interior of the apple is diseased, transmitting the apple to a special area, and if the interior of the apple is not diseased, entering the step S6;
s6, acquiring internal resonance sound wave data of the apples to be detected, judging the internal substance content of the apples by using the trained model, classifying the apples according to the substance content, and transmitting the screened apples to a designated area;
s7, resetting the quality screening interval to carry out additional screening on irrelevant apples in S4, and repeating the steps S2-S7 to finish the classification of all apples.
Preferably, the step S5 of acquiring the internal ultrasonic nondestructive data of the apples to be detected in the internal nondestructive testing area is to acquire the internal ultrasonic nondestructive data by using a non-contact air coupling technique and an ultrasonic penetration method. Air coupling ultrasonic detection technology: similar to the traditional water immersion method, the difference is that air is not water between the probe and the detected object. The incidence angle can be changed by changing the inclination angle of the probe, and meanwhile, ultrasonic waves of various modes can be excited, and focused ultrasonic waves can be formed more easily, so that internal ultrasonic nondestructive data can be acquired.
More preferably, the step S0 is:
labeling the appearance picture of the apple by using the quality interval in the step S1, and giving a label for training a convolutional neural network model (namely a CNN model) based on deep learning;
labeling is carried out by collecting internal ultrasonic nondestructive data and internal resonance sound wave data and using the quality interval in the step S1, and labels are given for training a convolutional neural network-cyclic neural network model (namely, a CNN-RNN model) based on deep learning.
More preferably, the training of the deep learning based convolutional neural network model comprises:
(1) preprocessing the acquired apple appearance image to obtain subimages, and randomly dividing the subimages into a training set, a verification set and a test set according to a certain proportion;
(2) inputting the data of the training set into a convolutional neural network for training to obtain a trained network model;
(3) inputting the data of the test set into the trained network model to obtain a preliminary test result;
(4) testing and evaluating the precision of the sample data set based on the trained model, debugging and optimizing the model parameters according to the evaluation result, and repeating the steps (1) to (3) until the model is stable to generate final model parameters;
the model adopts an SSIM loss function to describe the difference between the real value and the predicted value, and the model is debugged and optimized based on the difference, specifically:
Figure BDA0003201580740000031
wherein x and y represent two pictures, μx、μyDenotes the mean value of x, y, σx、σyDenotes the standard deviation, σ, of x, yxyDenotes the covariance of x, y, c1、c2Representing a constant and avoiding coefficient errors caused by denominator being zero.
Further preferably, the model described in step S3 is a CNN model (i.e., a convolutional neural network model).
Or more preferably, the training of the deep learning based convolutional neural network-cyclic neural network model comprises:
(1) preprocessing collected apple internal ultrasonic nondestructive data and internal resonance sound wave data, and randomly dividing the data into a training set, a verification set and a test set according to a certain proportion;
(2) inputting the data of the training set into a convolutional neural network-cyclic neural network for training to obtain a trained network model;
(3) inputting the data of the test set into the trained network model to obtain a preliminary test result;
(4) carrying out advancing test and precision evaluation on the sample data set based on the trained model, debugging and optimizing model parameters according to an evaluation result, and repeating the steps (1) to (3) until final model parameters are generated after the model is stabilized;
the model adopts a cyclic neural network to replace a convolutional neural network full-connection layer, and the output model is an LSTM model in the cyclic neural network.
Further preferably, the model described in steps S5 and S6 is a CNN-RNN model (i.e., a convolutional neural network-recurrent neural network model).
In another preferred embodiment, in step S3, the noise of salt and pepper in the picture is removed by using a median filtering method.
Wherein, the preferable formula of the median filtering method is as follows: cv2.media Blur (img, ksize),
where img is the image to be processed and ksize is the size of the filter kernel (height and width of its neighborhood image during the averaging process).
In another preferred embodiment, the quality screening interval setup in step S1 complies with GB/T10651-2008 and NYT 2316-.
The technical scheme provided by the embodiment of the invention has the following beneficial effects:
the invention directly processes, analyzes and judges the signals such as the sound waves obtained by nondestructive testing, and overcomes the defect of inaccurate detection result caused by the method that most of the market converts the obtained signals into pictures and finally analyzes and judges the pictures by optical testing.
The development of the apple nondestructive testing system based on machine learning can make up for the defects of the existing apple nondestructive testing system. The algorithm of deep learning is fused with a nondestructive testing technology, intellectualization is realized from the whole classification of placement, transmission and detection of apples, a large number of apples can be accurately detected in a high-speed and high-efficiency manner, the detection accuracy and efficiency are improved, and the detection flow is optimized. Not only can reduce the workload of the fruit grower, but also can improve the income of the fruit grower. And fruits with different grades can be provided according to the requirements of consumers, and certain effect is achieved on improving the position of domestic fruits.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings needed to be used in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.
FIG. 1 is a detection flowchart of example 1;
FIG. 2 is a CNN principle of operation;
FIG. 3 is a CNN-RNN principle of operation;
FIG. 4 is an image processing flow diagram;
fig. 5 is a schematic diagram of optical detection.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, embodiments of the present invention will be described in detail with reference to the accompanying drawings.
Example one
As shown in fig. 1-5, a machine learning-based apple nondestructive testing method includes:
and S0, collecting the appearance picture and the size data of the apple, the internal ultrasonic nondestructive data and the internal resonance sound wave data, and training the data to obtain a training model. The method specifically comprises the following steps:
labeling the appearance picture of the apple by using the quality interval in the step S1, and giving a label for training the convolutional neural network model based on deep learning.
Labeling by collecting internal ultrasonic nondestructive data and internal resonance sound wave data and using the quality interval in the step S1, and giving a label for training a convolutional neural network-cyclic neural network model based on deep learning.
The convolutional neural network model training process based on deep learning comprises the following steps:
(1) preprocessing the acquired apple appearance image to obtain subimages, and randomly dividing the subimages into a training set, a verification set and a test set according to a certain proportion;
(2) inputting the data of the training set into a convolutional neural network for training to obtain a trained network model;
(3) inputting the data of the test set into the trained network model to obtain a preliminary test result;
(4) testing and evaluating the precision of the sample data set based on the trained model, debugging and optimizing the model parameters according to the evaluation result, and repeating the steps (1) to (3) until the model is stable to generate final model parameters;
the model adopts an SSIM loss function to describe the difference between the real value and the predicted value, and the model is debugged and optimized based on the difference, specifically:
Figure BDA0003201580740000051
wherein x and y represent two pictures, μx、μyDenotes the mean value of x, y, σx、σyDenotes the standard deviation, σ, of x, yxyDenotes the covariance of x, y, c1、c2Representing a constant and avoiding coefficient errors caused by denominator being zero.
The convolutional neural network-cyclic neural network model training process based on deep learning comprises the following steps:
(1) preprocessing collected apple internal ultrasonic nondestructive data and internal resonance sound wave data, and randomly dividing the data into a training set, a verification set and a test set according to a certain proportion;
(2) inputting the data of the training set into a convolutional neural network-cyclic neural network for training to obtain a trained network model;
(3) inputting the data of the test set into the trained network model to obtain a preliminary test result;
(4) carrying out advancing test and precision evaluation on the sample data set based on the trained model, debugging and optimizing model parameters according to an evaluation result, and repeating the steps (1) to (3) until final model parameters are generated after the model is stabilized;
the model adopts a cyclic neural network to replace a convolutional neural network full-connection layer, and the output model is an LSTM model in the cyclic neural network.
S1, setting an apple quality screening interval; wherein the quality interval setting follows GB/T10651-2008 and NYT 2316-2013.
S2, conveying the apples to an appearance nondestructive detection area;
s3, obtaining an appearance picture of the apple to be detected, removing salt and pepper noise of the picture by using a median filtering method, identifying the appearance picture of the apple by using a trained CNN model, and determining the required apple according to the identification result;
the formula of the median filtering method is as follows: cv2.media Blur (img, ksize),
where img is the image to be processed and ksize is the size of the filter kernel (height and width of its neighborhood image during the averaging process).
S4, conveying the needed apples screened in the S3 to an internal nondestructive detection area, conveying irrelevant apples to an alternative selection area, and carrying out screening after parameters are reset;
s5, obtaining internal ultrasonic nondestructive data of the to-be-detected apple in the internal nondestructive detection area, and performing nondestructive judgment on the interior of the apple by using the trained CNN-RNN model, namely obtaining the internal ultrasonic nondestructive data by adopting a non-contact air coupling technology. Air coupling ultrasonic detection technology: similar to the traditional water immersion method, the difference is that air is not water between the probe and the detected object. The incidence angle can be changed by changing the inclination angle of the probe, and meanwhile, ultrasonic waves of various modes can be excited, and focused ultrasonic waves can be formed more easily, so that internal ultrasonic nondestructive data can be acquired. Determining that the interior of the mobile phone is not sick, if the interior of the mobile phone is sick, transmitting the mobile phone to a special area, and if the interior of the mobile phone is sick, entering step S6;
s6, obtaining internal resonance sound wave data of the apples to be detected, judging the internal substance content of the apples by using the trained CNN-RNN model, classifying according to the substance content, and transmitting the screened apples to a designated area;
s7, resetting the quality screening interval to carry out additional screening on irrelevant apples in S4, and repeating the steps S2-S7 to finish the classification of all apples.
In this embodiment, it is preferable that step S0 is to collect the appearance picture, size data, and internal ultrasound nondestructive data and internal resonance sound wave data of the apple by using the trinocular imaging acquisition system.
The trinocular imaging acquisition system comprises: the system comprises a data processing terminal, a data acquisition device, an apple detection device, an apple identification device and an information interaction device;
the data processing terminal is respectively connected with the data acquisition device, the apple detection device, the apple identification device and the information interaction device; the data acquisition device respectively acquires a visible light image and a depth image and inputs the visible light image and the depth image into the data processing terminal; the apple detection device acquires a depth image from the data processing terminal to carry out apple detection, and feeds back a detection result to the data processing terminal; the data processing terminal controls the apple identification module to identify the corresponding visible light image according to the apple detection result and feeds back the identification result to the data processing terminal; and the data processing terminal controls the data transmission of the information interaction device according to the apple detection and identification result.
The data processing terminal is provided with a data management module used for storing acquired images and identity information, and the data management module is respectively connected with the data processing terminal, the apple detection device and the apple identification device.
In the trinocular imaging acquisition system, the data acquisition device comprises an image acquisition device, and the image acquisition device comprises a visible light image acquisition module and a depth image acquisition module; the visible light image acquisition module and the depth image acquisition module are respectively connected with the data processing terminal.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.

Claims (10)

1. A machine learning-based apple nondestructive testing method is characterized by comprising the following steps:
s0, collecting appearance pictures and size data of the apples, internal ultrasonic nondestructive data and internal resonance sound wave data, and training the data to obtain a training model;
s1, setting an apple quality screening interval;
s2, conveying the apples to an appearance nondestructive detection area;
s3, obtaining an appearance picture of the apple to be detected, removing salt and pepper noise of the picture, identifying the appearance picture of the apple by using the trained model, and determining the required apple according to the identification result;
s4, conveying the needed apples screened in the S3 to an internal nondestructive detection area, conveying irrelevant apples to an alternative selection area, and carrying out screening after parameters are reset;
s5, obtaining internal ultrasonic nondestructive data of the to-be-detected apple in the internal nondestructive detection area, carrying out nondestructive judgment on the interior of the apple by using the trained model, determining that the interior of the apple is free of diseases, if the interior of the apple is diseased, transmitting the apple to a special area, and if the interior of the apple is not diseased, entering the step S6;
s6, acquiring internal resonance sound wave data of the apples to be detected, judging the internal substance content of the apples by using the trained model, classifying the apples according to the substance content, and transmitting the screened apples to a designated area;
s7, resetting the quality screening interval to carry out additional screening on irrelevant apples in S4, and repeating the steps S2-S7 to finish the classification of all apples.
2. The machine learning-based apple nondestructive testing method of claim 1, wherein the step S5 of obtaining the internal ultrasonic nondestructive data of the to-be-tested apples in the internal nondestructive testing area is to obtain the internal ultrasonic nondestructive data by a non-contact air coupling technique using an ultrasonic penetration method.
3. The apple nondestructive testing method based on machine learning of claim 2, wherein the step S0 is:
labeling the quality interval in the step S1 through the appearance picture of the apple, and giving a label for training a convolutional neural network model based on deep learning;
labeling by collecting internal ultrasonic nondestructive data and internal resonance sound wave data and using the quality interval in the step S1, and giving a label for training a convolutional neural network-cyclic neural network model based on deep learning.
4. The machine learning-based apple nondestructive testing method of claim 3, wherein the training of the deep learning-based convolutional neural network model comprises:
(1) preprocessing the acquired apple appearance image to obtain subimages, and randomly dividing the subimages into a training set, a verification set and a test set according to a certain proportion;
(2) inputting the data of the training set into a convolutional neural network for training to obtain a trained network model;
(3) inputting the data of the test set into the trained network model to obtain a preliminary test result;
(4) testing and evaluating the precision of the sample data set based on the trained model, debugging and optimizing the model parameters according to the evaluation result, and repeating the steps (1) to (3) until the model is stable to generate final model parameters;
the model adopts an SSIM loss function to describe the difference between the real value and the predicted value, and the model is debugged and optimized based on the difference, specifically:
Figure FDA0003201580730000021
wherein x and y represent two pictures, μx、μyDenotes the mean value of x, y, σx、σyDenotes the standard deviation, σ, of x, yxyDenotes the covariance of x, y, c1、c2Representing a constant and avoiding coefficient errors caused by denominator being zero.
5. The machine learning and machine learning-based apple nondestructive testing method of claim 4, wherein the model in step S3 is a CNN model.
6. The machine learning-based apple nondestructive testing method of claim 3, wherein the training of the deep learning-based convolutional neural network-cyclic neural network model comprises:
(1) preprocessing collected apple internal ultrasonic nondestructive data and internal resonance sound wave data, and randomly dividing the data into a training set, a verification set and a test set according to a certain proportion;
(2) inputting the data of the training set into a convolutional neural network-cyclic neural network for training to obtain a trained network model;
(3) inputting the data of the test set into the trained network model to obtain a preliminary test result;
(4) carrying out advancing test and precision evaluation on the sample data set based on the trained model, debugging and optimizing model parameters according to an evaluation result, and repeating the steps (1) to (3) until final model parameters are generated after the model is stabilized;
the model adopts a cyclic neural network to replace a convolutional neural network full-connection layer, and the output model is an LSTM model in the cyclic neural network.
7. The machine learning-based apple nondestructive testing method of claim 6, wherein the model in steps S5 and S6 is a CNN-RNN model.
8. The machine learning-based apple nondestructive testing method of claim 1, 2, 3 or 6, wherein in step S3, the picture salt and pepper noise is removed by median filtering.
9. The machine learning-based apple nondestructive testing method of claim 8, wherein the median filtering method formula is: cv2.media Blur (img, ksize),
where img is the image to be processed and ksize is the size of the filter kernel.
10. The method as claimed in claim 8, wherein the quality screening interval setup in step S1 complies with GB/T10651-2008 and NYT 2316-2013.
CN202110906221.5A 2021-08-09 2021-08-09 Apple nondestructive testing method based on machine learning Pending CN113624759A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110906221.5A CN113624759A (en) 2021-08-09 2021-08-09 Apple nondestructive testing method based on machine learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110906221.5A CN113624759A (en) 2021-08-09 2021-08-09 Apple nondestructive testing method based on machine learning

Publications (1)

Publication Number Publication Date
CN113624759A true CN113624759A (en) 2021-11-09

Family

ID=78383453

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110906221.5A Pending CN113624759A (en) 2021-08-09 2021-08-09 Apple nondestructive testing method based on machine learning

Country Status (1)

Country Link
CN (1) CN113624759A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115406967A (en) * 2022-11-01 2022-11-29 湖南木屋网络科技有限公司 Method, system, equipment and storage medium for judging fruit quality

Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5152401A (en) * 1989-10-13 1992-10-06 The United States Of America As Representd By The Secretary Of Agriculture Agricultural commodity condition measurement
JP2004202489A (en) * 2002-12-10 2004-07-22 Mitsui Mining & Smelting Co Ltd Article sorting system
JP2014173982A (en) * 2013-03-08 2014-09-22 Zensho Holdings Co Ltd Food inspection device and food inspection system
WO2018002841A1 (en) * 2016-06-29 2018-01-04 Ser.Mac S.R.L. An apparatus for detecting damaged fruit and vegetable products
US20180088084A1 (en) * 2016-09-28 2018-03-29 International Business Machines Corporation Food doneness monitor
CN110148122A (en) * 2019-05-17 2019-08-20 南京东奇智能制造研究院有限公司 Apple presentation quality stage division based on deep learning
CN111680738A (en) * 2020-06-03 2020-09-18 安徽中青检验检测有限公司 Screening device for apple quality detection and detection method thereof
CN112115752A (en) * 2019-06-21 2020-12-22 北京百度网讯科技有限公司 Method and device for training quality detection model and method and device for detecting quality
JP6830685B1 (en) * 2019-09-18 2021-02-17 Ａｓｓｅｓｔ株式会社 Apple quality estimation program and system
CN112560896A (en) * 2020-11-19 2021-03-26 安徽理工大学 Fruit quality screening and classifying system based on image processing

Patent Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5152401A (en) * 1989-10-13 1992-10-06 The United States Of America As Representd By The Secretary Of Agriculture Agricultural commodity condition measurement
JP2004202489A (en) * 2002-12-10 2004-07-22 Mitsui Mining & Smelting Co Ltd Article sorting system
JP2014173982A (en) * 2013-03-08 2014-09-22 Zensho Holdings Co Ltd Food inspection device and food inspection system
WO2018002841A1 (en) * 2016-06-29 2018-01-04 Ser.Mac S.R.L. An apparatus for detecting damaged fruit and vegetable products
US20180088084A1 (en) * 2016-09-28 2018-03-29 International Business Machines Corporation Food doneness monitor
CN110148122A (en) * 2019-05-17 2019-08-20 南京东奇智能制造研究院有限公司 Apple presentation quality stage division based on deep learning
CN112115752A (en) * 2019-06-21 2020-12-22 北京百度网讯科技有限公司 Method and device for training quality detection model and method and device for detecting quality
JP6830685B1 (en) * 2019-09-18 2021-02-17 Ａｓｓｅｓｔ株式会社 Apple quality estimation program and system
CN111680738A (en) * 2020-06-03 2020-09-18 安徽中青检验检测有限公司 Screening device for apple quality detection and detection method thereof
CN112560896A (en) * 2020-11-19 2021-03-26 安徽理工大学 Fruit quality screening and classifying system based on image processing

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115406967A (en) * 2022-11-01 2022-11-29 湖南木屋网络科技有限公司 Method, system, equipment and storage medium for judging fruit quality

Similar Documents

Publication Publication Date Title
CN110363746B (en) Ultrasonic nondestructive detection signal classification method based on convolutional neural network
CN102147402A (en) Machine vision technology based method for rapidly detecting egg freshness
CN110473179A (en) A kind of film surface defects detection method, system and equipment based on deep learning
CN110781889B (en) Deep learning-based nondestructive testing method for total sugar content in blueberry fruits
JP2011158373A (en) Method for creation of teacher data for use in automatic defect classification, and method and apparatus for automatic defect classification
CN202421024U (en) Broken rice rate on-line detection system
CN111652883B (en) Glass surface defect detection method based on deep learning
CN109829895A (en) A kind of AOI defect inspection method based on GAN
Sharma et al. Grain quality detection by using image processing for public distribution
CN102494977B (en) Method and system for detecting broken rice rate on line
CN111161237A (en) Fruit and vegetable surface quality detection method, storage medium and sorting device thereof
CN111160451A (en) Flexible material detection method and storage medium thereof
CN113252568A (en) Lens surface defect detection method, system, product and terminal based on machine vision
CN107543788A (en) A kind of urine erythrocyte abnormal rate detection method and system
CN114066848A (en) FPCA appearance defect visual inspection system
CN116559111A (en) Sorghum variety identification method based on hyperspectral imaging technology
CN111783616A (en) Data-driven self-learning-based nondestructive testing method
CN113624759A (en) Apple nondestructive testing method based on machine learning
CN117152152B (en) Production management system and method for detection kit
CN112697984A (en) Fruit defect nondestructive testing method based on neural network and fruit grading method
CN110991469A (en) Fruit soluble solid online detection method and system
CN102768212A (en) Fruit defect classification method based on compressed sensing
CN114898148B (en) Egg offset detection method and system based on deep learning
CN117237747B (en) Hardware defect classification and identification method based on artificial intelligence
CN117350967B (en) Optical detection-based wafer appearance defect detection system and method

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination