CN110570413B - Dual-automatic detection system for height matching of heels of soles - Google Patents
Dual-automatic detection system for height matching of heels of soles Download PDFInfo
- Publication number
- CN110570413B CN110570413B CN201910843983.8A CN201910843983A CN110570413B CN 110570413 B CN110570413 B CN 110570413B CN 201910843983 A CN201910843983 A CN 201910843983A CN 110570413 B CN110570413 B CN 110570413B
- Authority
- CN
- China
- Prior art keywords
- loss function
- channel
- target
- neural network
- deep neural
- 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.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01L—MEASURING FORCE, STRESS, TORQUE, WORK, MECHANICAL POWER, MECHANICAL EFFICIENCY, OR FLUID PRESSURE
- G01L5/00—Apparatus for, or methods of, measuring force, work, mechanical power, or torque, specially adapted for specific purposes
-
- 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
- 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
- G06N3/08—Learning methods
- G06N3/082—Learning methods modifying the architecture, e.g. adding, deleting or silencing nodes or connections
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
- G06Q10/06395—Quality analysis or management
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/04—Manufacturing
-
- 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
- G06T7/0004—Industrial image inspection
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Human Resources & Organizations (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Strategic Management (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Economics (AREA)
- Development Economics (AREA)
- Tourism & Hospitality (AREA)
- Quality & Reliability (AREA)
- General Business, Economics & Management (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Entrepreneurship & Innovation (AREA)
- Marketing (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Educational Administration (AREA)
- Primary Health Care (AREA)
- Manufacturing & Machinery (AREA)
- Game Theory and Decision Science (AREA)
- Operations Research (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a dual automatic detection system for the height of a heel of a shoe, which comprises: a first image acquisition device; a second image acquisition device; a first target image information input device; a second target image information input device; a first image recognition device; a second image recognition device; an image contrast analysis device. According to the invention, the image information of the high-low matching double heels is directly acquired through the first image acquisition device and the second image acquisition device, and after corresponding image identification is carried out, the identification result is obtained, the high-low matching double automatic detection of the heels of the soles is realized, in addition, partial areas of the heels clamped by the clamps are blocked by the clamps, the information of the partial areas cannot be acquired, the pressure information of the areas is acquired through the first pressure sensor array and the second pressure sensor array, the actual high-low matching double data of the areas is further judged, so that the heels detected by the high-low matching double automatic detection system of the heels of the soles have no detection dead angles, and the errors are further reduced.
Description
Technical Field
The invention belongs to the technical field of shoe product detection, and particularly relates to a dual-automatic detection system for height matching of heels of soles.
Background
Chinese patent application No. CN201520983772.1; the invention relates to a finished shoe testing machine which comprises a bending mechanism, a twisting mechanism and a clamp, wherein the twisting mechanism and the clamp are positioned on one side of the bending mechanism, the clamp is divided into a shoe head clamp and a heel clamp, the bending mechanism is arranged below the shoe head clamp and corresponds to the heel clamp in position, and the twisting mechanism is connected with one end of the heel clamp. The bending mechanism is provided with a load cell, the torsion mechanism is provided with a torque sensor, stress change can be detected, shoe performance can be analyzed, a processing scheme can be made in time, and the working state is stable. This finished shoe testing machine both can realize bending the test and can twist reverse the test, has realized multi-functionally, has reduced the input, and is very practical. The document in this application is a way of detecting shoes, but the heel detection still adopts a manual direct observation way to check, and especially in the detection process, the detection of constant pressure can not be realized. Chinese patent application No. cn201820657648.X; the polishing detection device for the heels comprises a plane for supporting an overall structure, wherein a supporting part for installing and detecting the heels is arranged on the plane, an inspection part for detecting the heels is arranged on the supporting part, the sampling detection device comprises an inspection part for rubbing the heels, a supporting part for installing the inspection part is arranged along the bottom side of the inspection part, the supporting part comprises a supporting surface for horizontally supporting the inspection part, a supporting platform which is horizontally arranged is arranged on the lower side of the supporting surface, a base which is integrated with the supporting platform is arranged on the lower portion of the supporting platform, a pressostat for positioning the heels is arranged on the plane and comprises a vertically arranged constant pressure shaft, a heel block for installing the heels is arranged at the bottom of the constant pressure shaft, and a limiting pin for limiting the constant pressure shaft is arranged in the middle position; the shoe heel has the advantages of simple integral structure, strong universality, high utilization rate, constant pressure at the bottom of the shoe heel, convenience in detection, high stability and convenience in use and popularization; although the defects of CN201520983772.1 are solved, the high-low matching double detection of the heel of the sole is not correspondingly explained.
In the shoe making industrial line in the prior art, shoes need to be subjected to quality detection after production is finished, and workers need to perform manual judgment when detecting whether the heel is consistent in shape; however, there is a certain error in this method, and a partial area of the clamp that holds the heel is blocked by the clamp, and information of this partial area cannot be collected, so that the high-low double detection of this partial area cannot be completed.
Therefore, how to provide a dual automatic detection system for detecting whether the heel of a shoe is high or low to complete the dual detection of the heel of the shoe in the area where information cannot be collected and reduce the error of manual judgment is an important technical problem to be solved urgently by those skilled in the art.
Disclosure of Invention
The invention aims to provide a dual-automatic detection system for height matching of heels of soles, which is used for solving the problems that in the prior art, on a shoe manufacturing industrial line, shoes need to carry out quality detection on the heels after production is finished, and when whether the heel shapes are consistent or not is detected, workers need to carry out manual judgment; however, this method has a problem that a certain error exists, and a partial area where the clamp holds the heel is blocked by the clamp, and information of the partial area cannot be acquired, so that the high-low double detection of the partial area cannot be completed.
In order to achieve the purpose, the technical scheme adopted by the invention is as follows:
two automatic check out system are joined in marriage to sole heel height includes:
the first image acquisition device is used for acquiring first target image information of a first heel with a matched pair in height;
the second image acquisition device is used for acquiring second target image information of a second heel with a matched pair in height;
the first target image information input device is used for inputting the first target image information into a first target model obtained by performing channel pruning on a deep neural network model by utilizing the characterization capability of a channel;
the second target image information input device is used for inputting the second target image information into a second target model obtained by performing channel pruning on the deep neural network model by utilizing the characterization capability of a channel;
the first image recognition device is used for carrying out classification processing on the first target image by utilizing a first auxiliary classifier in the first target model and obtaining a recognition result;
the second image recognition device is used for carrying out classification processing on the second target image by utilizing a second auxiliary classifier in the second target model and obtaining a recognition result;
the image comparison analysis device is used for comparing and analyzing the recognition result of the first image recognition device and the recognition result of the second image recognition device to obtain a matching rate so as to obtain a primary judgment result;
the first auxiliary classifier building device is used for inserting a batch normalization layer, a linear rectification layer and an average pooling layer into the deep neural network model to build a first auxiliary classifier;
the second auxiliary classifier building device is used for inserting a batch normalization layer, a linear rectification layer and an average pooling layer into the deep neural network model to build a second auxiliary classifier;
the first loss function constructing device is used for inserting an auxiliary loss function into the deep neural network model and forming a first target loss function with a reconstruction loss function of the deep neural network model;
the second loss function constructing device is used for inserting an auxiliary loss function into the deep neural network model and forming a second target loss function together with the reconstruction loss function of the deep neural network model;
the first training device is used for training the first auxiliary classifier by utilizing the first target loss function and combining the characterization capability of a channel to obtain a first target model;
the second training device is used for training the second auxiliary classifier by utilizing the second target loss function and combining the characterization capability of the channel to obtain a second target model;
further comprising:
the first pressure sensor array is used for acquiring first pressure information when a first heel with a pair of high and low heels is clamped by the clamp;
the second pressure sensor array is used for acquiring second pressure information when a second heel with a pair of high and low heels is clamped by the clamp;
the pressure data processing device is used for comparing and analyzing the first pressure information with preset first threshold pressure information and obtaining a corresponding analysis result, and is also used for comparing and analyzing the second pressure information with preset second threshold pressure information and obtaining a corresponding secondary judgment result;
and the heel matching judgment device is used for comparing the primary judgment result with the secondary judgment result and judging that the first heel and the second heel are in a high-low matching state when the primary judgment result is the same as the secondary judgment result.
Preferably, the first image acquisition device and the second image acquisition device both adopt a camera array.
Preferably, the auxiliary loss function inserted by the first loss function constructing device in the deep neural network model is a cross entropy loss function, and the auxiliary loss function inserted by the second loss function constructing device in the deep neural network model is also a cross entropy loss function.
Preferably, the first training device trains the first auxiliary classifier by using the first target loss function in combination with the characterization capability of the channel, so as to obtain a first target model; selecting a redundant channel to be pruned in a deep neural network model by utilizing the first target loss function and combining the characterization capability of the channel; in a deep neural network model, cutting off the redundant channel to obtain a first target model;
the second training device trains the second auxiliary classifier by using the second target loss function in combination with the characterization capability of the channel to obtain a second target model; selecting a redundant channel to be pruned from a deep neural network model by utilizing the first target loss function and combining the characterization capability of the channel; and in the deep neural network model, cutting off the redundant channel to obtain a second target model.
Preferably, when the first target loss function is utilized and the characterization capability of the channel is combined, a redundant channel to be pruned is selected in a deep neural network model; acquiring channel selection vectors representing the importance of each channel in a deep neural network model, optimizing the channel selection vectors and model parameters by using the first objective loss function, and determining the channel with the vector element of 0 in the optimized channel selection vectors as a redundant channel;
when the second target loss function is utilized and the characterization capability of the channel is combined, a redundant channel to be pruned is selected in the deep neural network model; and acquiring a channel selection vector representing the importance of each channel in the deep neural network model, optimizing the channel selection vector and the model parameters by using the second target loss function, and determining the channel with the vector element of 0 in the optimized channel selection vector as a redundant channel.
Preferably, when the channel selection vector and the model parameters are optimized by using the first objective loss function: randomly selecting a training sample, and enabling a first target loss function L (x, y) = LM (x, y) + ALS (x, y) to converge by using a random gradient descent algorithm and a greedy algorithm; wherein, LM is a reconstruction loss function, LS is a designated auxiliary loss function, x is a model parameter, and y is a channel selection vector; a is the weight of the specified auxiliary loss function; likewise, when optimizing the channel selection vector and the model parameters using the second objective loss function: randomly selecting a training sample, and enabling a second target loss function L (x, y) = LM (x, y) + ALS (x, y) to converge by using a random gradient descent algorithm and a greedy algorithm; wherein, LM is a reconstruction loss function, LS is a designated auxiliary loss function, x is a model parameter, and y is a channel selection vector; a is the weight of the specified auxiliary loss function.
The beneficial technical effects of the invention are as follows: on the shoemaking industry line, shoes need carry out quality testing to its heel after the production is accomplished, when detecting whether the heel shape height is unanimous, this application is joined in marriage the image information of two heels through first image acquisition device and second image acquisition device direct acquisition height, and carry out corresponding image identification after, obtain the identification result, the sole heel height has been realized and has been joined in marriage two automated inspection, and, the subregion of anchor clamps centre gripping heel is blocked by anchor clamps, can't gather this regional information, gather this regional pressure information through first pressure sensor array and second pressure sensor array, and then judge this regional actual height and join in marriage two data, make and join in marriage the sole heel that dual automated inspection system detects at this sole heel height and have not detected the dead angle, further reduced the error. In addition, the images are classified and identified through the deep neural network model, and the calculation amount of image identification can be greatly reduced.
Drawings
Fig. 1 is a schematic structural diagram of an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention are clearly and completely described below with reference to fig. 1 of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Example (b):
two automatic check out system are joined in marriage to sole heel height includes:
the first image acquisition device is used for acquiring first target image information of a first heel with a pair of high heels and low heels;
the second image acquisition device is used for acquiring second target image information of a second heel with double high heels and double low heels;
the first target image information input device is used for inputting the first target image information into a first target model obtained by performing channel pruning on the deep neural network model by utilizing the representation capability of a channel;
the second target image information input device is used for inputting the second target image information into a second target model obtained by performing channel pruning on the deep neural network model by utilizing the characterization capability of a channel;
the first image recognition device is used for carrying out classification processing on the first target image by utilizing a first auxiliary classifier in the first target model and obtaining a recognition result;
the second image recognition device is used for carrying out classification processing on the second target image by utilizing a second auxiliary classifier in the second target model and obtaining a recognition result;
the image comparison analysis device is used for comparing and analyzing the recognition result of the first image recognition device and the recognition result of the second image recognition device to obtain a matching rate, and further judging whether the height degree of the first heel is consistent with that of the second heel and obtaining a primary judgment result;
the first auxiliary classifier building device is used for inserting a batch normalization layer, a linear rectification layer and an average pooling layer into the deep neural network model to build a first auxiliary classifier;
the second auxiliary classifier building device is used for inserting a batch normalization layer, a linear rectification layer and an average pooling layer into the deep neural network model to build a second auxiliary classifier;
the first loss function constructing device is used for inserting an auxiliary loss function into the deep neural network model and forming a first target loss function with a reconstruction loss function of the deep neural network model;
the second loss function constructing device is used for inserting an auxiliary loss function into the deep neural network model and forming a second target loss function together with the reconstruction loss function of the deep neural network model;
the first training device is used for training the first auxiliary classifier by utilizing the first target loss function and combining the characterization capability of a channel to obtain a first target model;
the second training device is used for training the second auxiliary classifier by utilizing the second target loss function and combining the characterization capability of the channel to obtain a second target model;
further comprising:
the first pressure sensor array is used for acquiring first pressure information when a first heel with a pair of high and low heels is clamped by the clamp;
the second pressure sensor array is used for acquiring second pressure information when the second heel with the matched pairs in height is clamped by the clamp;
the pressure data processing device is used for comparing and analyzing the first pressure information and preset first threshold pressure information and obtaining a corresponding analysis result, and is also used for comparing and analyzing the second pressure information and preset second threshold pressure information and obtaining a corresponding secondary judgment result;
and the heel matching judgment device is used for comparing the primary judgment result with the secondary judgment result and judging that the first heel and the second heel are in a high-low matching state when the primary judgment result is the same as the secondary judgment result.
The working principle of the embodiment is briefly described as follows: on one hand, the first image acquisition device and the second image acquisition device can acquire most of video image information of a first heel and a second heel of a low-pair shoe which are not shielded by the heel clamping device, and identify whether the first heel and the second heel are of a pair of shoes with the same size according to a subsequent image identification mode, so that the automatic detection of the high-low pair of the heels can be quickly and accurately judged; on the other hand, the first pressure sensor array and the second pressure sensor array acquire first pressure information when a first heel of a high-low pair is clamped by the clamp and second pressure information when a second heel of the high-low pair is clamped by the clamp, the pressure data processing device compares the first pressure information with the second pressure information to judge whether the size of the part of the first heel clamped by the clamp is consistent with that of the part of the second heel clamped by the clamp, and outputs a secondary judgment result, and only when the primary judgment result and the secondary judgment result are high-low pair at the same time, the first heel and the second heel are judged to be in an accurate heel high-low pair state; through the detection of the two aspects, the actual condition of matching the height of the heel of the sole with the height of the heel can be accurately and completely detected; the sole heel that makes the detection does not have the detection dead angle, and the error in the detection of sole heel height pair further reduces.
Preferably, the first image acquisition device and the second image acquisition device both adopt a camera array, and can carry out comprehensive video image acquisition on a first heel and a second heel which are matched with each other in height.
Through above-mentioned scheme, camera array installs around first heel and second heel, carries out all-round image acquisition to the part that first heel and second heel expose, further reduces the error.
Preferably, the auxiliary loss function inserted by the first loss function constructing device in the deep neural network model is a cross entropy loss function, and the auxiliary loss function inserted by the second loss function constructing device in the deep neural network model is also a cross entropy loss function.
Preferably, the first training device trains the first auxiliary classifier by using the first target loss function in combination with the characterization capability of the channel, so as to obtain a first target model; selecting a redundant channel to be pruned from a deep neural network model by utilizing the first target loss function and combining the characterization capability of the channel; in a deep neural network model, cutting off the redundant channel to obtain a first target model;
the second training device trains the second auxiliary classifier by using the second target loss function in combination with the characterization capability of the channel to obtain a second target model; selecting a redundant channel to be pruned in a deep neural network model by utilizing the first target loss function and combining the characterization capability of the channel; and in the deep neural network model, cutting off the redundant channel to obtain a second target model.
According to the scheme, as the first training device and the second training device have the same principle, the first and second differences are abandoned, the channels with poor representation capability are removed from the target model, and then the target image is classified by using the auxiliary classifier in the target model, so that the identification result of the target image can be obtained; meanwhile, the channel removed by pruning is a channel with poor representation capability, so that the accuracy of the final identification result cannot be reduced by classifying and identifying the target image by adopting the deep neural network model after pruning; and through channel pruning, the calculated amount is greatly reduced, and after the redundant channel is pruned, the structure of the target model is more compact.
Preferably, when the first target loss function is utilized and the characterization capability of the channel is combined, a redundant channel to be pruned is selected in a deep neural network model; obtaining channel selection vectors representing the importance of each channel in a deep neural network model, optimizing the channel selection vectors and model parameters by using the first target loss function, and determining the channel with the vector element of 0 in the optimized channel selection vectors as a redundant channel;
when the second target loss function is utilized and the characterization capability of the channel is combined, a redundant channel to be pruned is selected in the deep neural network model; and acquiring a channel selection vector representing the importance of each channel in the deep neural network model, optimizing the channel selection vector and the model parameters by using the second target loss function, and determining the channel with the vector element of 0 in the optimized channel selection vector as a redundant channel.
Through the scheme, when the target loss function is used for optimizing the channel selection vector and the model parameters, in each step, all auxiliary loss functions are not used at the same time, but only two loss functions, namely the auxiliary loss function and the original loss function in the current step, are considered, namely, when the deep convolutional neural network is subjected to pruning optimization, the layers in the deep neural network model are taken as a unit, and layered optimization is carried out.
Preferably, when the channel selection vector and the model parameters are optimized by using the first objective loss function: randomly selecting a training sample, and enabling a first target loss function L (x, y) = LM (x, y) + ALS (x, y) to converge by using a random gradient descent algorithm and a greedy algorithm; wherein, LM is a reconstruction loss function, LS is a designated auxiliary loss function, x is a model parameter, and y is a channel selection vector; a is the weight of the specified auxiliary loss function; likewise, when optimizing the channel selection vector and the model parameters using the second objective loss function: randomly selecting a training sample, and enabling a second target loss function L (x, y) = LM (x, y) + ALS (x, y) to converge by using a random gradient descent algorithm and a greedy algorithm; wherein, LM is a reconstruction loss function, LS is a designated auxiliary loss function, x is a model parameter, and y is a channel selection vector; a is the weight of the specified auxiliary loss function.
Through the scheme, each channel selection vector can be selected by a greedy algorithm to obtain an important target channel in the input feature graph: in the deep neural network model, pruning is carried out in a mode of only reserving a target channel to obtain a pruning model. The loop iteration termination condition may be final loss function convergence, specifically, that is, the number of times of loop iteration is repeated reaches a preset value. The iteration times can be determined according to the actual precision requirement, namely when the precision requirement is higher, the iteration can be circularly performed for multiple times: when the accuracy requirement is low, the number of iterative loops may be correspondingly low.
In the description of the present invention, it is to be understood that the terms "counterclockwise", "clockwise", "longitudinal", "lateral", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate orientations or positional relationships based on those shown in the drawings, and are used merely for convenience of description of the present invention, and do not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and thus, should not be construed as limiting the present invention.
Claims (6)
1. Two automatic check out system are joined in marriage to sole heel height, its characterized in that includes:
the first image acquisition device is used for acquiring first target image information of a first heel with a matched pair in height;
the second image acquisition device is used for acquiring second target image information of a second heel with double high heels and double low heels;
the first target image information input device is used for inputting the first target image information into a first target model obtained by performing channel pruning on a deep neural network model by utilizing the characterization capability of a channel;
the second target image information input device is used for inputting the second target image information into a second target model obtained by performing channel pruning on the deep neural network model by utilizing the characterization capability of a channel;
the first image recognition device is used for carrying out classification processing on the first target image by utilizing a first auxiliary classifier in the first target model and obtaining a recognition result;
the second image recognition device is used for carrying out classification processing on the second target image by using a second auxiliary classifier in the second target model and obtaining a recognition result;
the image comparison analysis device is used for comparing and analyzing the recognition result of the first image recognition device and the recognition result of the second image recognition device to obtain a matching rate so as to obtain a primary judgment result;
the first auxiliary classifier building device is used for inserting a batch normalization layer, a linear rectification layer and an average pooling layer into the deep neural network model to build a first auxiliary classifier;
the second auxiliary classifier building device is used for inserting a batch normalization layer, a linear rectification layer and an average pooling layer into the deep neural network model to build a second auxiliary classifier;
the first loss function constructing device is used for inserting an auxiliary loss function into the deep neural network model and forming a first target loss function with a reconstruction loss function of the deep neural network model;
the second loss function constructing device is used for inserting an auxiliary loss function into the deep neural network model and forming a second target loss function together with the reconstruction loss function of the deep neural network model;
the first training device is used for training the first auxiliary classifier by utilizing the first target loss function and combining the characterization capability of a channel to obtain a first target model;
the second training device is used for training the second auxiliary classifier by utilizing the second target loss function and combining the characterization capability of the channel to obtain a second target model;
further comprising:
the first pressure sensor array is used for acquiring first pressure information when a first heel with a pair of high and low heels is clamped by the clamp;
the second pressure sensor array is used for acquiring second pressure information when a second heel with a pair of high and low heels is clamped by the clamp;
the pressure data processing device is used for comparing and analyzing the first pressure information and preset first threshold pressure information and obtaining a corresponding analysis result, and is also used for comparing and analyzing the second pressure information and preset second threshold pressure information and obtaining a corresponding secondary judgment result;
and the heel matching judgment device is used for comparing the primary judgment result with the secondary judgment result and judging that the first heel and the second heel are in a high-low matching state when the primary judgment result is the same as the secondary judgment result.
2. The system for automatically detecting the height of a heel of a shoe as claimed in claim 1, wherein the first image acquisition device and the second image acquisition device both adopt a camera array.
3. The system according to claim 1, wherein the auxiliary loss function inserted by the first loss function constructing device in the deep neural network model is a cross entropy loss function, and the auxiliary loss function inserted by the second loss function constructing device in the deep neural network model is also a cross entropy loss function.
4. The system according to claim 1, wherein the first training device trains the first auxiliary classifier by using the first target loss function in combination with the characterization capability of the channel to obtain a first target model; selecting a redundant channel to be pruned from a deep neural network model by utilizing the first target loss function and combining the characterization capability of the channel; in a deep neural network model, cutting off the redundant channel to obtain a first target model;
the second training device trains the second auxiliary classifier by using the second target loss function in combination with the characterization capability of the channel to obtain a second target model; selecting a redundant channel to be pruned in a deep neural network model by utilizing the second target loss function and combining the characterization capability of the channel; and in the deep neural network model, cutting off the redundant channel to obtain a second target model.
5. The dual automatic detection system for high-low matching of heels of claim 4, wherein when the first target loss function is utilized and the characterization capability of the channel is combined, a redundant channel to be pruned is selected in a deep neural network model; obtaining channel selection vectors representing the importance of each channel in a deep neural network model, optimizing the channel selection vectors and model parameters by using the first target loss function, and determining the channel with the vector element of 0 in the optimized channel selection vectors as a redundant channel;
selecting a redundant channel to be pruned in the deep neural network model by utilizing the second target loss function and combining the characterization capability of the channel; and acquiring a channel selection vector representing the importance of each channel in the deep neural network model, optimizing the channel selection vector and the model parameters by using the second target loss function, and determining the channel with the vector element of 0 in the optimized channel selection vector as a redundant channel.
6. The dual heel high-low fit automatic detection system according to claim 5, wherein when optimizing channel selection vectors and model parameters using the first objective loss function: randomly selecting a training sample, and enabling a first target loss function L (x, y) = LM (x, y) + ALS (x, y) to converge by using a random gradient descent algorithm and a greedy algorithm; wherein, LM is a reconstruction loss function, LS is a designated auxiliary loss function, x is a model parameter, and y is a channel selection vector; a is the weight of the specified auxiliary loss function; likewise, when optimizing the channel selection vector and the model parameters using the second objective loss function: randomly selecting a training sample, and converging a second target loss function L (x, y) = LM (x, y) + ALS (x, y) by using a random gradient descent algorithm and a greedy algorithm; wherein, LM is a reconstruction loss function, LS is a designated auxiliary loss function, x is a model parameter, and y is a channel selection vector; a is the weight of the specified auxiliary loss function.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910843983.8A CN110570413B (en) | 2019-09-06 | 2019-09-06 | Dual-automatic detection system for height matching of heels of soles |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910843983.8A CN110570413B (en) | 2019-09-06 | 2019-09-06 | Dual-automatic detection system for height matching of heels of soles |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110570413A CN110570413A (en) | 2019-12-13 |
CN110570413B true CN110570413B (en) | 2023-01-31 |
Family
ID=68778460
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910843983.8A Active CN110570413B (en) | 2019-09-06 | 2019-09-06 | Dual-automatic detection system for height matching of heels of soles |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110570413B (en) |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104207411A (en) * | 2014-09-26 | 2014-12-17 | 四川卡美迪鞋业有限公司 | Method for checking quality of molded shoes |
US9460557B1 (en) * | 2016-03-07 | 2016-10-04 | Bao Tran | Systems and methods for footwear fitting |
WO2017037504A1 (en) * | 2015-08-31 | 2017-03-09 | Ghahramani Barandagh Alireza | Smart shoes with analyser mechanism |
CN109393648A (en) * | 2018-10-13 | 2019-03-01 | 孙浩 | A kind of sole Weigh sensor production technology |
-
2019
- 2019-09-06 CN CN201910843983.8A patent/CN110570413B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104207411A (en) * | 2014-09-26 | 2014-12-17 | 四川卡美迪鞋业有限公司 | Method for checking quality of molded shoes |
WO2017037504A1 (en) * | 2015-08-31 | 2017-03-09 | Ghahramani Barandagh Alireza | Smart shoes with analyser mechanism |
US9460557B1 (en) * | 2016-03-07 | 2016-10-04 | Bao Tran | Systems and methods for footwear fitting |
CN109393648A (en) * | 2018-10-13 | 2019-03-01 | 孙浩 | A kind of sole Weigh sensor production technology |
Also Published As
Publication number | Publication date |
---|---|
CN110570413A (en) | 2019-12-13 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN114549522B (en) | Textile quality detection method based on target detection | |
TWI687937B (en) | Establishing method of chromosome abnormality detection model, chromosome abnormality detection system, and chromosome abnormality detection method | |
US20110188735A1 (en) | Method and device for defect inspection | |
CN110146798B (en) | Automatic analysis method and system for failure core particles | |
CN103954334B (en) | Fully automatic image pickup type water meter verification system and operating method thereof | |
CN114589631B (en) | Control system and method for numerical control relief grinding machine | |
CN103743698A (en) | Method for performing sensory evaluation on tea by adopting near infrared spectrum | |
CN112967942A (en) | Wafer testing method and device, computer storage medium and computer equipment | |
CN110570413B (en) | Dual-automatic detection system for height matching of heels of soles | |
CN106203476A (en) | A kind of pig's head tail wheel exterior feature recognition methods based on arest neighbors classification with fuzzy algorithmic approach | |
CN109685756A (en) | Image feature automatic identifier, system and method | |
CN113706455B (en) | Rapid detection method for damage of 330kV cable porcelain insulator sleeve | |
CN102914479A (en) | Automatic Brinell hardness testing method | |
CN117152152B (en) | Production management system and method for detection kit | |
KR101880489B1 (en) | shoe sole selection system and selection method | |
CN113269234A (en) | Connecting piece assembly detection method and system based on target detection | |
CN110322395B (en) | Part outline shape detection method and device based on image processing and affine transformation | |
CN116843743A (en) | Pomelo fruit volume measurement method based on three-view image contour fitting | |
CN103472064A (en) | Appraisal method of circular bright polished gemstone cut | |
CN106326927A (en) | Shoeprint new class detection method | |
CN116542910A (en) | Stamping part round hole positioning and deformation detection method based on sub-pixel edge extraction | |
CN102063533A (en) | Method for designing shoe tree according to pelma image analysis data | |
CN110111317A (en) | A kind of dispensing visual detection method for quality based on intelligent robot end | |
CN102509117A (en) | DTW (dynamic time warping) based method for automatically distinguishing abnormal measurements from plantar pressure measurement | |
CN113780488A (en) | Scratching and scratching defect detection method based on position polymerization degree |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |