CN105251707A - Defective goods eliminating and sorting device based on medical infusion visible impurity detecting system - Google Patents

Abstract

The invention discloses a defective goods eliminating and sorting device based on a medical infusion visible impurity detecting system. The device comprises a cutter, a swing plate, a driving device, a clamping cylinder and clamping fingers, wherein one end of the swing plate is mounted on a main shaft of the cutter; the driving device is arranged on the swing plate; the driving device comprises a motor, a motor mounting base, a cylinder mounting base and a lead screw; the clamping cylinder is mounted below the cylinder mounting base; the clamping fingers are arranged on the clamping cylinder; and the cutter and the motor are both controlled by a PLC control unit of a medical infusion visible impurity classifying, identifying and detecting system. The device is combined with an IDS-ELM algorithm to realize the identification and the classification of visual impurities and bubbles, can precisely classify and identify various defects, and can discard the defects with different types to different defective goods areas. The cutter is skillfully applied to solve the sorting problem which causes trouble during infusion for a long time in the prior art; and the device is simple in structure and easy to realize.

Description

Defective product removing and sorting device based on medical infusion visible foreign matter detection system

Technical Field

The invention belongs to the technical field of automation, and relates to a defective product removing and sorting device based on a medical infusion visible foreign matter detection system.

Background

In order to ensure the orderly production and production quality of the equipment, how to quickly and stably separate the detected defective products from the qualified products is a very challenging subject. According to the existing sorting method, defective products are rejected in a direct-shooting mode, a swinging mode and the like; according to actual debugging experience, when the container of splendid attire infusion greatly was the glass bottle, adopted direct rejection or the scheme is rejected to the oscillating, though simple structure, occupation space is little, but the dynamics of rejecting is difficult to be held and can produce the medicine bottle and bump, fall bottle or medicine bottle roll phenomenon, and this can directly lead to the shatter of medicine bottle body or the damage of medicine bottle encapsulation part. In the process of producing and manufacturing the large infusion solution, in order to facilitate the subsequent analysis of the reasons for generating the defective products, a device is needed to be designed to output the large infusion solution with different defects to different areas, and the stability of the defective product removing module is ensured.

Disclosure of Invention

In view of the above problems, the invention provides a defective product removing and sorting device based on a medical large infusion visible foreign matter detection system, which skillfully utilizes a divider and a servo driving device, and combines an IDS-ELM classification algorithm to distinguish large infusion defective products containing different types of visible foreign matters, and sorts corresponding defective products into different defective product areas according to different priorities of medicine bottles, so as to ensure accurate detection of the large infusion visible foreign matter detection system and prevent the medicine bottles from being damaged.

A defective product removing and sorting device based on a medical infusion visible foreign matter classification, identification and detection system comprises a divider, a swing plate, a driving device, a clamping cylinder and clamping fingers;

one end of the swinging plate is mounted on the main shaft of the divider, and the driving device is arranged on the swinging plate;

the driving device comprises a motor, a motor mounting seat, a cylinder mounting seat and a screw rod; the motor is fixed on the swing plate through the motor mounting seat, two ends of the screw rod are respectively mounted on a first bearing seat and a second bearing seat which are fixed on the swing plate through a first bearing and a second bearing, one end of the screw rod is driven by the motor through a coupler, and the other end of the screw rod is provided with a cylinder mounting seat;

the clamping cylinder is arranged below the cylinder mounting seat;

the clamping fingers are arranged on the clamping cylinder;

the divider and the motor are both controlled by a PLC control unit of the medical infusion visible foreign matter classification, identification and detection system.

The inner sides of the clamping fingers are provided with clamping pads.

The control instruction sent by the PLC control unit is sent according to the identification and classification result of the large infusion bottle to be detected, different classification results and different ascending heights of the main shaft of the divider are obtained, the large infusion bottle clamped by the clamping cylinder is placed to the corresponding foreign matter class, and the large infusion bottle is removed and classified;

the identification and classification process of the large infusion bottle is as follows:

step 1) continuously acquiring original images of a large infusion fluid being detected;

step 2), preprocessing an image;

performing Top-Hat morphological filtering processing on each frame of the large infusion image obtained in the step 1) to obtain a filtered image;

step 3), image segmentation;

carrying out image segmentation on the filtered image obtained in the step 2) by adopting a difference method to obtain a segmented image;

step 4), defect edge extraction;

extracting the defect edge in the large infusion image from the segmentation image obtained in the step 3);

the defects comprise visible foreign matters or air bubbles, the visible foreign matters comprise glass chips, hairs or floating objects, and the floating objects comprise rubber chips or fibers;

step 5), extracting a feature vector of the defect;

selecting characteristic parameters for describing the defects from the defect edges obtained in the step 4) to form characteristic vectors of the defects;

the characteristic parameters comprise shape characteristic parameters, gray characteristic parameters and motion characteristic parameters;

the shape characteristic parameters comprise a defect target area S, a defect target occupancy rate K and 7 geometric invariant moments of the defect, wherein the defect target occupancy rate refers to the ratio of the number of pixels of a defect target area to the minimum circumscribed rectangular area of the defect target area;

the gray characteristic parameters comprise a gray mean value of the defect target area and a gray standard deviation of the defect target area;

the motion characteristic parameters comprise the abscissa and the ordinate of the central point of the defect target;

step 6), visible foreign matters and bubbles are classified and identified;

the extracted feature vectors of the defect targets are classified and identified by using an ELM network model, if the classification and identification results of the defect targets are visible foreign matters, the corresponding large transfusion belongs to an unqualified product, and the defect types contained in the large transfusion are obtained according to the classification results;

the construction process of the ELM network model is as follows: firstly, setting 13 input nodes, 4 output nodes and 400 hidden layer nodes in an ELM algorithm network model, wherein the number range of the hidden layer nodes is 100-;

secondly, a training sample feature vector set of known defect types is selected and input into the ELM algorithm network model, and the ELM algorithm network model is trained to obtain a trained ELM network model.

The ELM network model applied in the step 6) is constructed by adopting an IDS-ELM algorithm, and the specific steps are as follows:

step 1: given a sample data set N (x)_i,t_i) Selecting a training set, x, from a given sample data set_iDenotes the ith sample, t_iRepresenting the classification result of the ith sample;

step 2: building an ELM network model f_L(x_i)；

Selecting the number L of hidden layer nodes of the initial network model as 400, the activation function of the hidden layer deviant as sigmoid, and randomly selecting a weight vector omega of an input layer connected with the hidden layer in (0,1)_jAnd offset b_j；

$f_L\left(x_i\right)=\underset{j=1}{\overset{L}{\Σ}}{\mathrm{\β}}_{j}g({\mathrm{\ω}}_j{x}_i+b_j)=o_i,1\≤j\≤L$

Wherein, ω is_j＝(ω_j1,ω_j2,...,ω_jn)^TRepresenting the connection weight vector between the jth hidden layer node and the input node, b_jOffset value representing the jth hidden layer node, β_j＝(β_j1,β_j2,...,β_jm)^TRepresenting a connection weight vector between the jth hidden layer node and the output node; o ° o_i＝(ο_i1,ο_i2,...,ο_im)^TOutputting a network model corresponding to the ith sample, g (x) is a sigmoid activation function, n takes a value of 13, and m takes a value of 4; o_iRepresenting the classification result output by the ith sample through the ELM network model;

step 3: let o_i＝t_iCalculating an output matrix H of the hidden layer of the ELM network model according to β ═ H⁺T, calculating a connection weight of an ELM network model hidden layer and an output layer, wherein T is an output matrix of the ELM network model, and calculating training precision train0 and training time0 of the ELM network model;

step 4: calculating the influence degree I of each hidden layer node_jAnd ordering according to the descending power to obtain ordered hidden layer nodes;

$I_j=\frac{D_j}{SD}$

$D_j=\frac{1}{L}\underset{i=1}{\overset{N}{\Σ}}\left|g_j\left(x_i\right)\right|\left|\right|{\mathrm{\β}}_j\left|\right|+\frac{a}{N}\left|\right|{\mathrm{\ω}}_j\left|\right|,SD=\underset{j=1}{\overset{L}{\Σ}}{D}_j$

wherein, g_j(x_i)＝g(ω_jx_i+b_j) I is more than or equal to 1 and less than or equal to N, j is more than or equal to 1 and less than or equal to L, and a is an influence factor of the weight vector of the input layer, a ∈ (0, 1);

step 5: carrying out first pruning on the ELM network model;

selecting front lambda hidden layer nodes from the sorted hidden layer nodes obtained from step4, wherein lambda belongs to [1,5], and lambda is a positive integer; deleting the ELM from the ELM network model, calculating the training precision train1 of the ELM network model after the first pruning, recalculating the influence degree of each hidden layer node in the ELM network model after the first pruning according to step4, and sorting the nodes according to the power reduction;

the pruning coefficient is calculated η and, is a rounded up symbol;

step 6: carrying out second pruning on the ELM network model;

with eta · λ as the number of hidden nodes for the second pruning, selecting front eta · λ hidden layer nodes from the hidden layer nodes in descending order obtained from step5 to prune the first pruned ELM network model obtained from step5, and calculating the training accuracy train2 of the second pruned ELM network model;

step 7: retrieving the hidden layer node with the maximum influence degree in the second pruning operation, adding the hidden layer node into the ELM network model obtained by step6 again, and calculating the training precision train3 of the updated ELM network model;

step 8: cutting off a hidden layer node with the minimum influence degree from the ELM network model obtained from step6 to obtain the training precision of the updated ELM network model, namely train 4;

step 9: determining the number of final hidden nodes of the ELM network model as L', the training precision as train, and taking the train as max (train2, train3, train4), wherein the training time is time:

$L^{\′}=\left\{\begin{array}{c}L-(\mathrm{\λ}+1)-1,train=train4\\ L-(\mathrm{\λ}+1),train=train2\\ L-(\mathrm{\λ}+1)+1,train=train3\end{array}\right.$

step10, calculating weight matrix β 'of the network hidden layer and the output layer connection by using minimum norm least square solution of the contradictory linear equation set, β ═ H'). T, and updating weight vector omega of the input layer connection hidden layer_jAnd offset b_jObtaining a trained final ELM network model;

and H' is an output layer matrix of the final ELM network model.

In the morphological filtering process, a 7 × 7 circular template is selected as a structural element to perform top-hat morphological filtering on the original image.

In the step 3), an inter-frame difference method based on the maximum information entropy is adopted for image segmentation, and the specific steps are as follows:

firstly, carrying out difference operation on a continuously acquired sequence image to obtain a difference image;

next, the image binarization threshold T0 after difference is calculated:

respectively calculating the total number N of pixels in the target region of the image to be detected²Proportion p of pixel points with gray level i_iThe distribution of the background and foreign matter gray values is calculated using the following two equations:

$\frac{p_{i+1}}{1-Z_s},\frac{p_{i+2}}{1-Z_s}...\frac{p_M}{1-Z_s}$

$\frac{p_1}{Z_s},\frac{p_2}{Z_s},...\frac{p_i}{Z_s}$

wherein,m represents the maximum value of the gray level i, the information entropies H (A) of the background and the target,H (b) can be calculated by the following two formulas, respectively:

$H\left(A\right)=\underset{j=s+1}{\overset{M}{\Σ}}\frac{p_j}{1-Z_S}ln\frac{p_j}{1-Z_S}$

$H\left(B\right)=\underset{j=1}{\overset{M}{\Σ}}\frac{p_j}{Z_S}ln\frac{p_j}{Z_S}$

the total information entropy of the image to be detected can be obtained as phi(s) ═ H (A) + H (B) according to the two formulas, and when the phi(s) is enabled to be the maximum value, a differential image binarization threshold value T0 is obtained;

and finally, performing binarization processing on the differentiated image by using the differentiated image binarization threshold value T0 according to the following formula, performing AND operation on each pixel point in the obtained binarization image to obtain a symmetrical differential binary image, and finishing image segmentation:

$B{\left(x,y\right)}_{k_{n_o},k_{n_o+1}}=\left\{\begin{array}{cc}blackground& D(k_{n_o},k_{n_o+1})<T0\\ 255& D(k_{n_o},k_{n_o+1})\&GreaterEqual;T0\end{array}\right.$

extracting the defect edge in the large infusion image from the segmentation image obtained in the step 3) by adopting an SUSAN algorithm, and specifically comprising the following steps:

traversing each pixel of a target area in the symmetrical differential binary image by using a mask, comparing a gray value of a central pixel point of the mask with each pixel point in the mask area, recording the pixel points of which the gray difference value is smaller than a set gray difference value threshold, and forming the recorded pixel points into a USAN area;

the pixel values of all the pixel points except the central point in the mask are calculated by the following formula:

$C(r,r_0)=\left\{\begin{array}{cc}1& |I\left(r\right)-I\left(r_0\right)|\&le;t\\ 0& |I\left(r\right)-I\left(r_0\right)|>t\end{array}\right.$

r₀is the location of the image kernel, r represents the location of the rest of the points in the template, I (r)₀) Pixel values representing core points of the image, i (r) pixel values representing other points in the image template;

the USAN value of the mask area is then calculated using the following formula:

$n(x_0,y_0)=\underset{(x,y)\&NotEqual;\left(x_o,y_o\right)}{\&Sigma;}c(x,y)$

wherein (x)₀,y₀) The representation is the center point of the current mask, (x, y) represents the pixel points of the current mask except the center point, n is the number of pixels in a USAN region, then the pixel points are compared with a preset USAN threshold value, suspicious feature points are obtained by using the following formula, the feature points are compared with the gray values of 8 points in other adjacent regions by taking the feature points as the center, and the maximum pixel points are reserved as final edge points:

$R(x_0,y_0)=\left\{\begin{array}{cc}g-n(x_0,y_0),& n(x_0,y_0)<g\\ 0,& n(x_0,y_0)\&GreaterEqual;g\end{array}\right.$

wherein g ═ n_max(ii)/2 is the USAN threshold, n_maxIs the maximum value of n, 3/4 of the mask is taken.

For images affected by noise, the SUSAN threshold lower limit takes 2-10 pixels.

A classification recognition detection system for visible foreign matters and bubbles in 250ml medical infusion solution comprises a mechanical execution unit, a PLC control unit, a visual imaging unit and an industrial personal computer;

the mechanical execution unit is controlled by a PLC control unit, and the PLC control unit is connected with the industrial personal computer;

the visual imaging unit comprises an industrial camera and a light source, wherein the industrial camera is connected with the industrial personal computer, and the light source is controlled by the PLC control unit;

the mechanical execution unit comprises a large infusion input channel to be detected, a guide wheel disc, a main wheel disc, a bottle grabbing mechanical arm, a bottle twisting mechanism, a defective product removing and sorting device and a qualified product output channel;

a plurality of large infusion bottle stations are uniformly arranged on the main wheel disc, and each station is provided with a bottle grabbing mechanical arm and a bottle twisting mechanism;

the bottle grabbing mechanical hand comprises a bottle grabbing cylinder 17, a bottle grabbing cylinder mounting plate 18, a pull rod 21, a rotary sleeve 20, an angular contact bearing 29, a pull block 26 and a clamping hand 22;

the clamping handle 22 is fixed with the rotating sleeve 20 through a first pin shaft 23, and the clamping handle 22 freely makes circular motion by taking the first pin shaft 23 as a circle center; the rotating sleeve 20 is fixed on a rotating wheel bearing of the main wheel disc through a first angular contact bearing 28; the pull rod 21 is connected with two symmetrically distributed pull blocks 26 through a third pin shaft 25; the pulling block 26 is connected with the clamping hand 22 through a second pin shaft 27;

a second angular contact bearing 29 is mounted at the upper end of the pull rod and connected with a connector, the connector is connected with the cylinder 17 through a floating connector, and the bottle grabbing cylinder 17 is fixed on the bottle grabbing cylinder mounting plate;

the clamping hand 22 is provided with a clamping hand buffer block 24.

The large infusion input channel to be detected is connected with the leading-in wheel disc, the leading-out wheel disc is connected with the genuine product output channel, and the leading-in wheel disc and the leading-out wheel disc are respectively arranged on two sides of the main wheel disc and are in transmission connection with the main wheel disc;

the defective product removing and sorting device is arranged on one side of the quality product output channel and is controlled by the PLC control unit.

Defective product rejecting and sorting device: according to the priority, the large transfusion liquid containing different types of defects is output to different areas through different defective product output channels, so that secondary detection and analysis of causes of defective products are facilitated.

In order to avoid the damage of the packaging part of a large infusion bottle, the air inlet of the clamping cylinder is adjusted and controlled through a precise pressure regulating valve, the main shaft of the four-station (adjusted to be double stations) divider can move up and down, and the up-and-down movement stroke and the rotating station of the divider are adjusted through assembly and adjustment, so that different types of inferior-quality products can be conveyed to different inferior-quality product output channels from the vicinity of the outlet of the qualified product output channel. In order to facilitate the customer to identify the large infusion mixed with different foreign matters, the swing plate is provided with a set of driving device. Under the coordination of the photoelectric sensors arranged on the swinging plate, the driving device is controlled by corresponding PLC instructions to output different types of inferior-quality products according to 'priority'.

The method comprises the steps of shooting continuous multi-frame images by using a visual imaging unit, preprocessing and segmenting the images, extracting features after edge extraction, classifying and identifying the extracted feature vector set describing visible foreign matters and bubbles by using an IDS-ELM algorithm, and outputting the detected large transfusion liquid through a defective product removing and sorting device and a qualified product output channel.

The visual imaging unit comprises an industrial camera and a light source, wherein the industrial camera is connected with the industrial personal computer, and the light source is controlled by the PLC control unit.

Advantageous effects

The invention provides a defective product removing and sorting device based on a medical infusion visible foreign matter detection system, which comprises a divider, a swing plate, a driving device, a clamping cylinder and clamping fingers, wherein the divider is arranged on the upper portion of the swing plate; one end of the swinging plate is mounted on the main shaft of the divider, and the driving device is arranged on the swinging plate; the driving device comprises a motor, a motor mounting seat, a cylinder mounting seat and a screw rod; the clamping cylinder is arranged below the cylinder mounting seat; the clamping fingers are arranged on the clamping cylinder; the divider and the motor are both controlled by a PLC control unit of the medical infusion visible foreign matter classification, identification and detection system. The sorting of the large transfusion defective products is realized by the partial motion of the upper part of the bottle neck of the large transfusion medicine bottle.

Its advantages are the following:

(1) the clamping pad arranged on the clamping finger can ensure that the large infusion can be clamped stably and effectively protect the packaging part of the large infusion bottle mouth;

(2) the device can effectively distinguish different types of inferior-quality products in the process of removing the inferior-quality products. The vertical direction of the swinging plate is provided with three stations, (the specific number of stations can be determined according to the types of defective products distinguished by a large infusion manufacturer, and the applicability of the device is enhanced) which respectively correspond to three types of defective products, namely hair, floaters and glass chips, a corresponding photoelectric sensor is arranged on the swinging plate on each corresponding station, when the defective products are clamped by clamping fingers, a driving device drives a large infusion liquid to convey the defective products of different types to different stations along the direction of the swinging plate, then a divider main shaft is lifted up and drives the clamped defective products to move upwards, then the divider main shaft rotates 90 degrees, and then the defective products are put down. Therefore, different inferior-quality products can be output to different positions according to the priority, the inferior-quality products can be effectively distinguished while being removed, and the subsequent analysis on the causes of the inferior-quality products is facilitated. The divider ingeniously solves the sorting problem which puzzles large infusion for a long time in the prior art, and is simple in structure and easy to realize.

(3) The device effectively avoids the damage to the large transfusion container caused by cylinder direct-hit and other removing modes.

(4) The preprocessing image is obtained by adopting mathematical form filtering, and the algorithm can be realized in parallel through hardware, so that the processing speed is greatly improved. For the segmentation of the large infusion image, the defect that the bottle side wall possibly has difference and the detection effect of simple sequence image difference on the tiny foreign object target is poor is well overcome by adopting the interframe difference algorithm based on the maximum information entropy, the signal-to-noise ratio of the output image is obviously improved, the edge of the visible foreign object and the bubble is extracted by adopting the SUSAN algorithm, and the difficulty of the subsequent feature extraction algorithm is greatly simplified. The characteristic parameters of visible foreign matters and bubbles are analyzed and researched, and a characteristic descriptor for describing defects is constructed, so that the image processing time is reduced, and the real-time performance and the robustness of the algorithm are greatly improved. The IDS-ELM algorithm is adopted to realize the classification and identification of visible foreign matters and bubbles, and can classify and identify various defects with high precision and eliminate the defects containing different types to different defective areas.

Description of the drawings:

FIG. 1 is an overall structure of a defective product removing and sorting apparatus;

FIG. 2 is an overall configuration diagram of the servo drive apparatus;

FIG. 3 is an explanatory view of a structure of a servo drive device;

FIG. 4 is a schematic view of the overall structure of the detection system;

FIG. 5 is a schematic view of a bottle grasping mechanism, (a) is a schematic view of grasping, and (b) is a schematic view of releasing;

description of reference numerals: 1-large infusion medicine bottle, 2-clamping cylinder, 3-clamping pad, 4-clamping finger, 5-divider, 6-driving device, 7-swinging plate, 8-motor, 9-motor mounting seat, 10-coupler, 11-first bearing, 12-first bearing seat, 13-lead screw, 14-cylinder mounting seat, 15-second bearing, 16-second bearing seat, 17-medicine bottle grabbing manipulator cylinder, 18-cylinder mounting plate, 19-connector, 20-rotating sleeve, 21-pull rod, 22-gripper, 23-first pin shaft, 24-gripper buffer block, 25-third pin shaft, 26-pull block, 27-second pin shaft, 28-first corner contact bearing, 29-second corner contact bearing, 30-a floating joint, 31-a medicine bottle grabbing mechanical arm, 32-a bottle twisting mechanism, 33-an output channel, 34-a defective product removing and sorting device, 35-a guide-out wheel disc, 36-a guide-in wheel disc and 37-an input channel.

DETAILED DESCRIPTION OF EMBODIMENT (S) OF INVENTION

The invention will be further described with reference to the following figures and examples.

As shown in fig. 1-3, a defective product rejecting and sorting device based on a medical infusion visible foreign matter classification, identification and detection system comprises a divider, a swing plate, a driving device, a clamping cylinder and a clamping finger;

one end of the swinging plate is mounted on the main shaft of the divider, and the driving device is arranged on the swinging plate;

the driving device comprises a motor, a motor mounting seat, a cylinder mounting seat and a lead screw; the motor is fixed on the swing plate through the motor mounting seat, two ends of the screw rod are respectively mounted on a first bearing seat and a second bearing seat which are fixed on the swing plate through a first bearing and a second bearing, one end of the screw rod is driven by the motor through a coupler, and the other end of the screw rod is provided with a cylinder mounting seat;

the clamping cylinder is arranged below the cylinder mounting seat;

the clamping fingers are arranged on the clamping cylinder;

the divider and the motor are both controlled by a PLC control unit of the medical infusion visible foreign matter classification, identification and detection system.

The inner sides of the clamping fingers are provided with clamping pads.

The specific sorting process is as follows:

1) continuously acquiring multiframe original images of the large transfusion liquid being detected, then carrying out Top-Hat morphological filtering on the images, obtaining segmented images from the filtered images by adopting an interframe difference method based on maximum information entropy, extracting edges of visible foreign matters and bubbles in the large transfusion liquid images by adopting a SUSAN algorithm, selecting characteristic parameters for describing the shapes, gray scales and motion of the visible foreign matters (hair, floaters and glass chips) and the bubbles by researching and analyzing the characteristics of defects, and identifying and classifying the visible foreign matters and the bubbles by using an IDS-ELM algorithm on the extracted characteristic vectors for describing the visible foreign matters and the bubbles so as to distinguish defective products containing different types of foreign matters;

2) sorting various defective products by a defective product removing and sorting device:

before step 2) is specified, the "priority" of the transport is defined according to the market demand: if the detected large transfusion liquid contains different types of foreign matters, determining a final output channel of the large transfusion liquid according to the quantity of the foreign matters, for example, if the detected large transfusion liquid contains 1 hair and 2 glass chips, carrying the large transfusion liquid to the glass chip output channel by the device; if the detected large infusion fluid contains the same amount of different foreign matters, the output channel is selected in order of the glass cullet, the hair, and the float, and if the detected large infusion fluid contains 1 fiber and 1 hair, for example, the defective item removing mechanism conveys the large infusion fluid to the hair output channel.

The control instruction sent by the PLC control unit is sent according to the identification and classification result of the large infusion bottle to be detected, different classification results and different ascending heights of the main shaft of the divider are obtained, the large infusion bottle clamped by the clamping cylinder is placed to the corresponding foreign matter class, and the large infusion bottle is removed and classified;

the identification and classification process of the large infusion bottle is as follows:

step 1) continuously acquiring original images of a large infusion fluid being detected;

step 2), preprocessing an image;

performing Top-Hat morphological filtering processing on each frame of the large infusion image obtained in the step 1) to obtain a filtered image;

selecting a 7 multiplied by 7 circular template as a structural element to perform top hat morphological filtering on an original image;

step 3), image segmentation;

carrying out image segmentation on the filtered image obtained in the step 2) by adopting a difference method to obtain a segmented image;

extracting the defect edge in the large infusion image from the segmentation image obtained in the step 3);

the defects comprise visible foreign matters or air bubbles, the visible foreign matters comprise glass chips, hairs or floating objects, and the floating objects comprise rubber chips or fibers;

the method comprises the following steps of adopting an interframe difference method based on the maximum information entropy to carry out image segmentation:

firstly, carrying out difference operation on a continuously acquired sequence image to obtain a difference image;

next, the image binarization threshold T0 after difference is calculated:

respectively calculating the total number N of pixels in the target region of the image to be detected²Proportion p of pixel points with gray level i_iThe distribution of the background and foreign matter gray values is calculated using the following two equations:

$\frac{p_{i+1}}{1-Z_s},\frac{p_{i+2}}{1-Z_s}...\frac{p_M}{1-Z_s}$

$\frac{p_1}{Z_s},\frac{p_2}{Z_s},...\frac{p_i}{Z_s}$

wherein,m represents the maximum value of the gray level i, the information entropies H (A), H (B) of the background and the target can be calculated by the following two formulas respectively:

$H\left(A\right)=\underset{j=s+1}{\overset{M}{\&Sigma;}}\frac{p_j}{1-Z_S}ln\frac{p_j}{1-Z_S}$

$H\left(B\right)=\underset{j=1}{\overset{M}{\&Sigma;}}\frac{p_j}{Z_S}ln\frac{p_j}{Z_S}$

the total information entropy of the image to be detected can be obtained as phi(s) ═ H (A) + H (B) according to the two formulas, and when the phi(s) is enabled to be the maximum value, a differential image binarization threshold value T0 is obtained;

and finally, performing binarization processing on the differentiated image by using the differentiated image binarization threshold value T0 according to the following formula, performing AND operation on each pixel point in the obtained binarization image to obtain a symmetrical differential binary image, and finishing image segmentation:

$B{\left(x,y\right)}_{k_{n_o},k_{n_o+1}}=\left\{\begin{array}{cc}blackground& D(k_{n_o},k_{n_o+1})<T0\\ 255& D(k_{n_o},k_{n_o+1})\&GreaterEqual;T0\end{array}\right.$

step 4), defect edge extraction;

extracting the defect edge in the large infusion image from the segmentation image obtained in the step 3) by adopting an SUSAN algorithm, and specifically comprising the following steps:

traversing each pixel of a target area in the symmetrical differential binary image by using a mask, comparing a gray value of a central pixel point of the mask with each pixel point in the mask area, recording the pixel points of which the gray difference value is smaller than a set gray difference value threshold, and forming the recorded pixel points into a USAN area;

the pixel values of all the pixel points except the central point in the mask are calculated by the following formula:

$C(r,r_0)=\left\{\begin{array}{cc}1& |I\left(r\right)-I\left(r_0\right)|\&le;t\\ 0& |I\left(r\right)-I\left(r_0\right)|>t\end{array}\right.$

r₀is the location of the image kernel, r represents the location of the rest of the points in the template, I (r)₀) Pixel values representing core points of the image, i (r) pixel values representing other points in the image template;

the USAN value of the mask area is then calculated using the following formula:

$n(x_0,y_0)=\underset{(x,y)\&NotEqual;\left(x_o,y_o\right)}{\&Sigma;}c(x,y)$

wherein (x)₀,y₀) The representation is the center point of the current mask, (x, y) represents the pixel points of the current mask except the center point, n is the number of pixels in a USAN region, then the pixel points are compared with a preset USAN threshold value, suspicious feature points are obtained by using the following formula, the feature points are compared with the gray values of 8 points in other adjacent regions by taking the feature points as the center, and the maximum pixel points are reserved as final edge points:

$R(x_0,y_0)=\left\{\begin{array}{cc}g-n(x_0,y_0),& n(x_0,y_0)<g\\ 0,& n(x_0,y_0)\&GreaterEqual;g\end{array}\right.$

wherein g ═ n_max(ii)/2 is the USAN threshold, n_maxIs the maximum value of n, 3/4 of the mask is taken.

For images with noise effects, the USAN lower threshold value takes 2-10 pixels.

Step 5), extracting a feature vector of the defect;

selecting characteristic parameters for describing the defects from the defect edges obtained in the step 4) to form characteristic vectors of the defects;

the characteristic parameters comprise shape characteristic parameters, gray characteristic parameters and motion characteristic parameters;

the shape characteristic parameters comprise a defect target area S, a defect target occupancy rate K and 7 geometric invariant moments of the defect, wherein the defect target occupancy rate refers to the ratio of the number of pixels of a defect target area to the minimum circumscribed rectangular area of the defect target area;

the gray characteristic parameters comprise a gray mean value of the defect target area and a gray standard deviation of the defect target area;

the motion characteristic parameters comprise the abscissa and the ordinate of the central point of the defect target;

step 6), visible foreign matters and bubbles are classified and identified;

the extracted feature vectors of the defect targets are classified and identified by using an ELM network model, if the classification and identification results of the defect targets are visible foreign matters, the corresponding large transfusion belongs to an unqualified product, and the defect types contained in the large transfusion are obtained according to the classification results;

the construction process of the ELM network model is as follows: firstly, setting 13 input nodes, 4 output nodes and 400 hidden layer nodes in an ELM algorithm network model, wherein the number range of the hidden layer nodes is 100-;

secondly, a training sample feature vector set of known defect types is selected and input into the ELM algorithm network model, and the ELM algorithm network model is trained to obtain a trained ELM network model.

The ELM network model applied in the step 6) is constructed by adopting an IDS-ELM algorithm, and the specific steps are as follows:

step 1: given a sample data setN(x_i,t_i) Selecting a training set, x, from a given sample data set_iDenotes the ith sample, t_iRepresenting the classification result of the ith sample;

step 2: building an ELM network model f_L(x_i)；

Selecting the number L of hidden layer nodes of the initial network model as 400, the activation function of the hidden layer deviant as sigmoid, and randomly selecting a weight vector omega of an input layer connected with the hidden layer in (0,1)_jAnd offset b_j；

$f_L\left(x_i\right)=\underset{j=1}{\overset{L}{\&Sigma;}}{\mathrm{\&beta;}}_{j}g({\mathrm{\&omega;}}_j{x}_i+b_j)=o_i,1\&le;j\&le;L$

Wherein, ω is_j＝(ω_j1,ω_j2,...,ω_jn)^TRepresenting the connection weight vector between the jth hidden layer node and the input node, b_jOffset value representing the jth hidden layer node, β_j＝(β_j1,β_j2,...,β_jm)^TRepresenting a connection weight vector between the jth hidden layer node and the output node; o ° o_i＝(ο_i1,ο_i2,...,ο_im)^TOutputting a network model corresponding to the ith sample, g (x) is a sigmoid activation function, n takes a value of 13, and m takes a value of 4; o_iRepresenting the classification result output by the ith sample through the ELM network model;

step 3: let o_i＝t_iCalculating an output matrix H of the hidden layer of the ELM network model according to β ═ H⁺T, calculating a connection weight of an ELM network model hidden layer and an output layer, wherein T is an output matrix of the ELM network model, and calculating training precision train0 and training time0 of the ELM network model;

step 4: calculating the influence degree I of each hidden layer node_jAnd ordering according to the descending power to obtain ordered hidden layer nodes;

$I_j=\frac{D_j}{SD}$

$D_j=\frac{1}{L}\underset{i=1}{\overset{N}{\&Sigma;}}\left|g_j\left(x_i\right)\right|\left|\right|{\mathrm{\&beta;}}_j\left|\right|+\frac{a}{N}\left|\right|{\mathrm{\&omega;}}_j\left|\right|,SD=\underset{j=1}{\overset{L}{\&Sigma;}}{D}_j$

wherein, g_j(x_i)＝g(ω_jx_i+b_j) I is more than or equal to 1 and less than or equal to N, j is more than or equal to 1 and less than or equal to L, and a is an influence factor of the weight vector of the input layer, a ∈ (0, 1);

step 5: carrying out first pruning on the ELM network model;

selecting front lambda hidden layer nodes from the sorted hidden layer nodes obtained from step4, wherein lambda belongs to [1,5], and lambda is a positive integer; deleting the ELM from the ELM network model, calculating the training precision train1 of the ELM network model after the first pruning, recalculating the influence degree of each hidden layer node in the ELM network model after the first pruning according to step4, and sorting the nodes according to the power reduction;

the pruning coefficient is calculated η and, is a rounded up symbol;

step 6: carrying out second pruning on the ELM network model;

with eta · λ as the number of hidden nodes for the second pruning, selecting front eta · λ hidden layer nodes from the hidden layer nodes in descending order obtained from step5 to prune the first pruned ELM network model obtained from step5, and calculating the training accuracy train2 of the second pruned ELM network model;

step 7: retrieving the hidden layer node with the maximum influence degree in the second pruning operation, adding the hidden layer node into the ELM network model obtained by step6 again, and calculating the training precision train3 of the updated ELM network model;

step 8: cutting off a hidden layer node with the minimum influence degree from the ELM network model obtained from step6 to obtain the training precision of the updated ELM network model, namely train 4;

step 9: determining the number of final hidden nodes of the ELM network model as L', the training precision as train, and taking the train as max (train2, train3, train4), wherein the training time is time:

$L^{\&prime;}=\left\{\begin{array}{c}L-(\mathrm{\&lambda;}+1)-1,train=train4\\ L-(\mathrm{\&lambda;}+1),train=train2\\ L-(\mathrm{\&lambda;}+1)+1,train=train3\end{array}\right.$

step10, calculating weight matrix β 'of the network hidden layer and the output layer connection by using minimum norm least square solution of the contradictory linear equation set, β ═ H'). T, and updating weight vector omega of the input layer connection hidden layer_jAnd offset b_jObtaining a trained final ELM network model;

and H' is an output layer matrix of the final ELM network model.

In order to avoid the damage of the packaging part of a large infusion bottle, the air inlet of the clamping cylinder needs to be adjusted and controlled through a precise pressure regulating valve, the main shaft of the four-station (double-station) divider can move up and down, and the up-and-down movement stroke and the rotating station of the divider need to be adjusted through assembly and adjustment, so that different types of inferior-quality products can be conveyed to different inferior-quality product output channels from the vicinity of the outlet of the qualified product output channel. In order to facilitate the customer to identify the large infusion liquid mixed with different foreign matters, the swing plate is provided with a set of servo carrying device 6, and the servo driving components are controlled by corresponding PLC instructions under the coordination of the photoelectric sensors to output different types of inferior-quality products according to 'priority'.

A new classification, identification and detection system for visible foreign matters in medical infusion is constructed by adopting the sorting mechanism, and comprises a mechanical execution unit, a PLC (programmable logic controller) control unit, a visual imaging unit and an industrial personal computer, as shown in figure 4;

the mechanical execution unit is controlled by a PLC control unit, and the PLC control unit is connected with the industrial personal computer;

the visual imaging unit comprises an industrial camera and a light source, wherein the industrial camera is connected with the industrial personal computer, and the light source is controlled by the PLC control unit;

the mechanical execution unit comprises a large infusion input channel to be detected, a guide wheel disc, a main wheel disc, a bottle grabbing mechanical arm, a bottle twisting mechanism, a defective product removing and sorting device and a qualified product output channel;

a plurality of large infusion bottle stations are uniformly arranged on the main wheel disc, and each station is provided with a bottle grabbing mechanical arm and a bottle twisting mechanism;

the large infusion input channel to be detected is connected with the leading-in wheel disc, the leading-out wheel disc is connected with the genuine product output channel, and the leading-in wheel disc and the leading-out wheel disc are respectively arranged on two sides of the main wheel disc and are in transmission connection with the main wheel disc;

the defective product removing and sorting device is arranged on one side of the quality product output channel and is controlled by the PLC control unit.

As shown in fig. 5, the bottle grabbing manipulator comprises a bottle grabbing cylinder (17), a bottle grabbing cylinder mounting plate (18), a pull rod (21), a rotary sleeve (20), an angular contact bearing (29), a pull block (26) and a clamping hand (22);

the clamping hand (22) is fixed with the rotating sleeve (20) through a first pin shaft (23), and the clamping hand (22) freely performs circular motion by taking the first pin shaft (23) as a circle center; the rotating sleeve (20) is fixed on a rotating wheel bearing of the main wheel disc through a first angle contact bearing (28); the pull rod (21) is connected with two symmetrically distributed pull blocks (26) through a third pin shaft (25); the pulling block (26) is connected with the clamping hand (22) through a second pin shaft (27);

a second angular contact bearing (29) is mounted at the upper end of the pull rod and connected with a connector, the connector is connected with the cylinder (17) through a floating connector, and the bottle grabbing cylinder (17) is fixed on the bottle grabbing cylinder mounting plate;

the clamping hand (22) is provided with a clamping hand buffer block (24).

The PLC is utilized to control the servo motor, so that different types of inferior-quality products can be quickly, accurately and stably removed to different areas, secondary detection is realized, and the follow-up analysis on the reasons of the inferior-quality products is facilitated.

Claims (10)

1. A defective product removing and sorting device based on a medical infusion visible foreign matter classification, identification and detection system is characterized by comprising a divider (5), a swing plate (7), a driving device (6), a clamping cylinder (2), a clamping pad (3) and clamping fingers (4);

one end of the swinging plate is mounted on the main shaft of the divider, and the driving device is arranged on the swinging plate;

the driving device comprises a motor (8), a motor mounting seat (9), a cylinder mounting seat (14) and a screw rod (13); the motor is fixed on the swing plate through a motor mounting seat, two ends of the screw rod are respectively mounted on a first bearing seat (12) and a second bearing seat (16) which are fixed on the swing plate through a first bearing (11) and a second bearing (15), one end of the screw rod is driven by the motor through a coupler (10), and the other end of the screw rod is provided with a cylinder mounting seat;

the clamping cylinder is arranged below the cylinder mounting seat;

the clamping fingers are arranged on the clamping cylinder;

the divider and the motor are both controlled by a PLC control unit of the medical infusion visible foreign matter classification, identification and detection system.

2. The device of claim 1, wherein the inside of the gripping fingers are provided with gripping pads.

3. The device according to claim 1 or 2, wherein the control command sent by the PLC control unit is sent according to the identification and classification result of the large infusion bottle to be detected, different classification results are obtained, the rotation angle of the main shaft of the divider is different from the lifting height, the large infusion bottle clamped by the clamping cylinder is placed to the corresponding foreign matter class, and the defective large infusion bottle is removed and classified;

the identification and classification process of the large infusion bottle is as follows:

step 1) continuously acquiring original images of a large infusion fluid being detected;

step 2), preprocessing an image;

performing Top-Hat morphological filtering processing on each frame of the large infusion image obtained in the step 1) to obtain a filtered image;

step 3), image segmentation;

carrying out image segmentation on the filtered image obtained in the step 2) by adopting a difference method to obtain a segmented image;

step 4), defect edge extraction;

extracting the defect edge in the large infusion image from the segmentation image obtained in the step 3);

the defects comprise visible foreign matters or air bubbles, the visible foreign matters comprise glass chips, hairs or floating objects, and the floating objects comprise rubber chips or fibers;

step 5), extracting a feature vector of the defect;

selecting characteristic parameters for describing the defects from the defect edges obtained in the step 4) to form characteristic vectors of the defects;

the characteristic parameters comprise shape characteristic parameters, gray characteristic parameters and motion characteristic parameters;

the shape characteristic parameters comprise a defect target area S, a defect target occupancy rate K and 7 geometric invariant moments of the defect, wherein the defect target occupancy rate refers to the ratio of the number of pixels of a defect target area to the minimum circumscribed rectangular area of the defect target area;

the gray characteristic parameters comprise a gray mean value of the defect target area and a gray standard deviation of the defect target area;

the motion characteristic parameters comprise the abscissa and the ordinate of the central point of the defect target;

step 6), visible foreign matters and bubbles are classified and identified;

the extracted feature vectors of the defect targets are classified and identified by using an ELM network model, if the classification and identification results of the defect targets are visible foreign matters, the corresponding large transfusion belongs to an unqualified product, and the defect types contained in the large transfusion are obtained according to the classification results;

the construction process of the ELM network model is as follows: firstly, setting 13 input nodes, 4 output nodes and 400 hidden layer nodes in an ELM algorithm network model, wherein the number range of the hidden layer nodes is 100-;

secondly, a training sample feature vector set of known defect types is selected and input into the ELM algorithm network model, and the ELM algorithm network model is trained to obtain a trained ELM network model.

4. The apparatus of claim 3, wherein the ELM network model applied in step 6) is constructed by using IDS-ELM algorithm, and comprises the following steps:

step 1: given a sample data set N (x)_i,t_i) From a given number of samplesSelecting training set, x, from data set_iDenotes the ith sample, t_iRepresenting the classification result of the ith sample;

step 2: building an ELM network model f_L(x_i)；

Selecting the number L of hidden layer nodes of the initial network model as 400, the activation function of the hidden layer deviant as sigmoid, and randomly selecting a weight vector omega of an input layer connected with the hidden layer in (0,1)_jAnd offset b_j；

$f_L\left(x_i\right)=\underset{j=1}{\overset{L}{\&Sigma;}}{\mathrm{\&beta;}}_{j}g({\mathrm{\&omega;}}_j{x}_i+b_j)=o_i,1\&le;j\&le;L$

Wherein, ω is_j＝(ω_j1,ω_j2,...,ω_jn)^TRepresenting the connection weight vector between the jth hidden layer node and the input node, b_jOffset value representing the jth hidden layer node, β_j＝(β_j1,β_j2,...,β_jm)^TRepresenting a connection weight vector between the jth hidden layer node and the output node; o ° o_i＝(ο_i1,ο_i2,...,ο_im)^TOutputting a network model corresponding to the ith sample, g (x) is a sigmoid activation function, n takes a value of 13, and m takes a value of 4; o_iRepresenting the classification result output by the ith sample through the ELM network model;

step 3: let o_i＝t_iCalculating an output matrix H of the hidden layer of the ELM network model according to β ═ H⁺T, calculating a connection weight of an ELM network model hidden layer and an output layer, wherein T is an output matrix of the ELM network model, and calculating training precision train0 and training time0 of the ELM network model;

step 4: calculating the influence degree I of each hidden layer node_jAnd ordering according to the descending power to obtain ordered hidden layer nodes;

$I_j=\frac{D_j}{SD}$

$D_j=\frac{1}{L}\underset{i=1}{\overset{N}{\&Sigma;}}\left|g_j\left(x_i\right)\right|\left|\right|{\mathrm{\&beta;}}_j\left|\right|+\frac{a}{N}\left|\right|{\mathrm{\&omega;}}_j\left|\right|,SD=\underset{j=1}{\overset{L}{\&Sigma;}}{D}_j$

wherein, g_j(x_i)＝g(ω_jx_i+b_j) I is more than or equal to 1 and less than or equal to N, j is more than or equal to 1 and less than or equal to L, and a is an influence factor of the weight vector of the input layer, a ∈ (0, 1);

step 5: carrying out first pruning on the ELM network model;

selecting front lambda hidden layer nodes from the sorted hidden layer nodes obtained from step4, wherein lambda belongs to [1,5], and lambda is a positive integer; deleting the ELM from the ELM network model, calculating the training precision train1 of the ELM network model after the first pruning, recalculating the influence degree of each hidden layer node in the ELM network model after the first pruning according to step4, and sorting the nodes according to the power reduction;

the pruning coefficient is calculated η and,is a rounded up symbol;

step 6: carrying out second pruning on the ELM network model;

with eta · λ as the number of hidden nodes for the second pruning, selecting front eta · λ hidden layer nodes from the hidden layer nodes in descending order obtained from step5 to prune the first pruned ELM network model obtained from step5, and calculating the training accuracy train2 of the second pruned ELM network model;

step 7: retrieving the hidden layer node with the maximum influence degree in the second pruning operation, adding the hidden layer node into the ELM network model obtained by step6 again, and calculating the training precision train3 of the updated ELM network model;

step 8: cutting off a hidden layer node with the minimum influence degree from the ELM network model obtained from step6 to obtain the training precision of the updated ELM network model, namely train 4;

step 9: determining the number of final hidden nodes of the ELM network model as L', the training precision as train, and taking the train as max (train2, train3, train4), wherein the training time is time:

$L^{\&prime;}=\left\{\begin{array}{cc}L-(\mathrm{\&lambda;}+1)-1,& train=train4\\ L-(\mathrm{\&lambda;}+1),& train=train2\\ L-(\mathrm{\&lambda;}+1)+1,& train=train3\end{array}\right.$

step10, calculating weight matrix β 'of the network hidden layer and the output layer connection by using minimum norm least square solution of the contradictory linear equation set, β ═ H'). T, and updating weight vector omega of the input layer connection hidden layer_jAnd offset b_jObtaining a trained final ELM network model;

and H' is an output layer matrix of the final ELM network model.

5. The apparatus of claim 4, wherein in the morphological filtering process, a 7 x 7 circular template is used as a structural element to perform top-hat morphological filtering on the original image.

6. The apparatus according to claim 5, wherein the image segmentation is performed by using an inter-frame difference method based on maximum entropy in step 3), and the specific steps are as follows:

firstly, carrying out difference operation on a continuously acquired sequence image to obtain a difference image;

next, the image binarization threshold T0 after difference is calculated:

respectively calculating the total number N of pixels in the target region of the image to be detected²Proportion p of pixel points with gray level i_iThe distribution of the background and foreign matter gray values is calculated using the following two equations:

$\frac{p_{i+1}}{1-Z_s},\frac{p_{i+2}}{1-Z_s}...\frac{p_M}{1-Z_s}$

$\frac{p_1}{Z_s},\frac{p_2}{Z_s},...\frac{p_i}{Z_s}$

wherein,m represents the maximum value of the gray level i, the information entropies H (A), H (B) of the background and the target can be calculated by the following two formulas respectively:

$H\left(A\right)=\underset{j=s+1}{\overset{M}{\&Sigma;}}\frac{p_j}{1-Z_S}ln\frac{p_j}{1-Z_S}$

$H\left(B\right)=\underset{j=1}{\overset{M}{\&Sigma;}}\frac{p_j}{Z_S}ln\frac{p_j}{Z_S}$

the total information entropy of the image to be detected can be obtained as phi(s) ═ H (A) + H (B) according to the two formulas, and when the phi(s) is enabled to be the maximum value, a differential image binarization threshold value T0 is obtained;

and finally, performing binarization processing on the differentiated image by using the differentiated image binarization threshold value T0 according to the following formula, performing AND operation on each pixel point in the obtained binarization image to obtain a symmetrical differential binary image, and finishing image segmentation:

$B{\left(x,y\right)}_{k_{n_o},k_{n_o+1}}=\left\{\begin{array}{cc}blackground& D(k_{n_o},k_{n_o+1})<T0\\ 255& D(k_{n_o},k_{n_o+1})\&GreaterEqual;T0\end{array}\right..$

7. the device according to claim 6, wherein a SUSAN algorithm is used to extract the defect edge in the large infusion image from the segmentation image obtained in step 3), and the specific steps are as follows:

traversing each pixel of a target area in the symmetrical differential binary image by using a mask, comparing a gray value of a central pixel point of the mask with each pixel point in the mask area, recording the pixel points of which the gray difference value is smaller than a set gray difference value threshold, and forming the recorded pixel points into a USAN area;

the pixel values of all the pixel points except the central point in the mask are calculated by the following formula:

$C(r,r_0)=\left\{\begin{array}{cc}1& |I\left(r\right)-I\left(r_0\right)|\&le;t\\ 0& |I\left(r\right)-I\left(r_0\right)|>t\end{array}\right.$

r₀is the location of the image kernel, r represents the location of the rest of the points in the template, I (r)₀) Pixel values representing core points of the image, i (r) pixel values representing other points in the image template;

the USAN value of the mask area is then calculated using the following formula:

$n(x_0,y_0)=\underset{(x,y)\&NotEqual;(x_o,y_o)}{\&Sigma;}c(x,y)$

wherein (x)₀,y₀) The representation is the current mask center point, (x)Y) represents the pixel points of the current mask except the central point, n is the number of pixels in the USAN region, then the pixel points are compared with a preset USAN threshold value, suspicious characteristic points are obtained by using the following formula, the characteristic points are used as the center, the characteristic points are compared with 8 point gray values in other neighborhoods, and the maximum pixel points are reserved as final edge points:

$R(x_0,y_0)=\left\{\begin{array}{cc}g-n\left(x_0,y_0\right),& n\left(x_0,y_0\right)<g\\ 0,& n\left(x_0,y_0\right)\&GreaterEqual;g\end{array}\right.$

wherein g ═ n_max(ii)/2 is the USAN threshold, n_maxIs the maximum value of n, 3/4 of the mask is taken.

8. The apparatus of claim 7, wherein the SUSAN threshold lower limit value is 2-10 pixels for noisy images.

9. A classification recognition detection system for visible foreign matters and bubbles in 250ml medical infusion solution is characterized by comprising a mechanical execution unit, a PLC control unit, a visual imaging unit and an industrial personal computer;

the mechanical execution unit is controlled by a PLC control unit, and the PLC control unit is connected with the industrial personal computer;

the visual imaging unit comprises an industrial camera and a light source, wherein the industrial camera is connected with the industrial personal computer, and the light source is controlled by the PLC control unit;

the mechanical execution unit comprises a large transfusion input channel 37 to be detected, a guide-in wheel disc 36, a guide-out wheel disc 35, a bottle grabbing manipulator 31, a bottle twisting mechanism 32, a defective product removing and sorting device 34 and a genuine product output channel 33;

a plurality of large infusion bottle stations are uniformly arranged on the main wheel disc, and each station is provided with a bottle grabbing mechanical arm and a bottle twisting mechanism;

the large infusion input channel to be detected is connected with the leading-in wheel disc, the leading-out wheel disc is connected with the genuine product output channel, and the leading-in wheel disc and the leading-out wheel disc are respectively arranged on two sides of the main wheel disc and are in transmission connection with the main wheel disc;

the defective product removing and sorting device is arranged on one side of the quality product output channel and is controlled by the PLC control unit.

10. The system according to claim 9, wherein the bottle grabbing mechanical hand comprises a bottle grabbing cylinder (17), a bottle grabbing cylinder mounting plate (18), a pull rod (21), a rotating sleeve (20), an angular contact bearing (29), a pull block (26) and a clamping hand (22);

the clamping hand (22) is fixed with the rotating sleeve (20) through a first pin shaft (23), and the clamping hand (22) freely performs circular motion by taking the first pin shaft (23) as a circle center; the rotating sleeve (20) is fixed on a rotating wheel bearing of the main wheel disc through a first angle contact bearing (28); the pull rod (21) is connected with two symmetrically distributed pull blocks (26) through a third pin shaft (25); the pulling block (26) is connected with the clamping hand (22) through a second pin shaft (27);

a second angular contact bearing (29) is mounted at the upper end of the pull rod and connected with a connector, the connector is connected with the cylinder (17) through a floating connector, and the bottle grabbing cylinder (17) is fixed on the bottle grabbing cylinder mounting plate;

the clamping hand (22) is provided with a clamping hand buffer block (24).