CN118135339B - Monitoring management method and system for chilli food production and processing - Google Patents
Monitoring management method and system for chilli food production and processingInfo
- Publication number
- CN118135339B CN118135339B CN202410544877.0A CN202410544877A CN118135339B CN 118135339 B CN118135339 B CN 118135339B CN 202410544877 A CN202410544877 A CN 202410544877A CN 118135339 B CN118135339 B CN 118135339B
- Authority
- CN
- China
- Prior art keywords
- pepper
- black spot
- image
- node
- black
- 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
- 238000012545 processing Methods 0.000 title claims abstract description 35
- 238000012544 monitoring process Methods 0.000 title claims abstract description 25
- 235000013305 food Nutrition 0.000 title claims abstract description 23
- 238000004519 manufacturing process Methods 0.000 title claims abstract description 22
- 238000007726 management method Methods 0.000 title claims description 10
- 240000004160 Capsicum annuum Species 0.000 title abstract 2
- 235000002566 Capsicum Nutrition 0.000 claims abstract description 176
- 239000006002 Pepper Substances 0.000 claims abstract description 129
- 241000722363 Piper Species 0.000 claims abstract description 129
- 235000016761 Piper aduncum Nutrition 0.000 claims abstract description 129
- 235000017804 Piper guineense Nutrition 0.000 claims abstract description 129
- 235000008184 Piper nigrum Nutrition 0.000 claims abstract description 129
- 206010027146 Melanoderma Diseases 0.000 claims abstract description 110
- 241000758706 Piperaceae Species 0.000 claims abstract description 30
- 238000013145 classification model Methods 0.000 claims abstract description 30
- 238000000034 method Methods 0.000 claims abstract description 30
- 239000012535 impurity Substances 0.000 claims abstract description 24
- 238000002372 labelling Methods 0.000 claims abstract description 8
- 238000003709 image segmentation Methods 0.000 claims abstract description 5
- 238000001514 detection method Methods 0.000 claims abstract description 4
- 240000008574 Capsicum frutescens Species 0.000 claims description 39
- 239000011159 matrix material Substances 0.000 claims description 27
- 238000012549 training Methods 0.000 claims description 21
- 239000001390 capsicum minimum Substances 0.000 claims description 17
- 238000004364 calculation method Methods 0.000 claims description 9
- 238000011176 pooling Methods 0.000 claims description 9
- 238000005096 rolling process Methods 0.000 claims description 8
- 238000003708 edge detection Methods 0.000 claims description 7
- 238000000547 structure data Methods 0.000 claims description 5
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 238000003860 storage Methods 0.000 claims description 3
- 239000002994 raw material Substances 0.000 abstract description 8
- 239000002699 waste material Substances 0.000 abstract description 3
- 230000006870 function Effects 0.000 description 6
- 235000015067 sauces Nutrition 0.000 description 5
- 238000004140 cleaning Methods 0.000 description 4
- 230000001788 irregular Effects 0.000 description 4
- 238000003475 lamination Methods 0.000 description 4
- 238000012795 verification Methods 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 238000009826 distribution Methods 0.000 description 2
- 239000013049 sediment Substances 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 241000764238 Isis Species 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 235000009508 confectionery Nutrition 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 235000011194 food seasoning agent Nutrition 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
Abstract
The invention relates to the technical field of image processing, and discloses a monitoring and managing method and a managing system for producing and processing chilli food, wherein the method comprises the following steps: step 101, acquiring pepper information, wherein the pepper information comprises pepper variety information, growing environment information and pepper appearance images; 102, performing image segmentation on the acquired pepper appearance images to obtain all pepper individual appearance images, and performing pixel-level labeling on black spot areas in all the pepper individual appearance images; step 103, modeling based on the marked pepper individual appearance image to generate a black spot classification model; step 104, deploying the black spot classification model into an automatic detection system of a production line; the monitoring and managing method can accurately identify the pepper individuals with impurities on the surfaces and still can be used, and avoids the situation that the pepper individuals are mistakenly removed together with peppers with black spots formed by decay, thereby reducing the waste of raw materials and improving the utilization rate of the raw materials.
Description
Technical Field
The invention relates to the field of image processing, in particular to a monitoring and managing method and a monitoring and managing system for pepper food production and processing.
Background
The chilli sauce is a salty, sweet and spicy fermented seasoning, the chilli sauce is produced and processed by taking chilli as a main raw material, in the current full-automatic chilli sauce processing procedure, all chilli needs to be cleaned before processing in order to ensure the food safety and sanitary quality of chilli sauce products, but the source of the chilli is numerous, the quality of the chilli is different, the cleaning degree is different, after cleaning, a plurality of chilli surfaces still have 'black spots', and the chilli with the 'black spots' on the surfaces is usually screened out directly in order to ensure the quality of the chilli sauce products, so that a great amount of chilli raw material loss and material cost increase are caused.
The peppers with the 'black spots' on the surfaces contain a large amount of peppers with impurities adhered on the surfaces, the peppers can be used as raw materials after being further cleaned, the peppers with the 'black spots' adhered on the surfaces are directly cleaned in the prior art, the peppers with the impurities adhered on the surfaces are easily caused to be removed by error screening, unnecessary waste is generated, and the cost of enterprises is increased.
Disclosure of Invention
The invention provides a monitoring and managing method and a managing system for producing and processing chilli food, which solve the technical problems in the related art.
The invention provides a monitoring and managing method for producing and processing chilli food, which comprises the following steps:
Step 101, acquiring pepper information, wherein the pepper information comprises pepper variety information, growing environment information and pepper appearance images;
102, performing image segmentation on the acquired pepper appearance images to obtain all pepper individual appearance images, performing pixel-level labeling on black spot areas in all the pepper individual appearance images, and marking each pixel of the black spot areas in the pepper individual appearance images as one of black spots belonging to peppers or adhering impurities;
step 103, modeling based on the marked pepper individual appearance image to generate a black spot classification model;
and 104, deploying the black spot classification model into an automatic detection system of a production line, and selecting a corresponding processing mode of the pepper individuals by using the classification result of the pepper individual appearance image output by the black spot classification model.
In step 102, the marking rule is that the types of the pixels of the black spot area in the appearance image of the pepper individual are summarized, if the number of the types of the pixels of the black spot area marked as the black spots belonging to the pepper is more than half, the black spot area is marked as the black spot belonging to the pepper, otherwise, the black spot area is marked as the adhered impurity.
In step 102, the method for processing the appearance image of the pepper individual to obtain the black spot region category comprises the following steps:
step one, carrying out gray conversion on the collected pepper appearance image, and converting a color image into a gray image;
Secondly, performing edge detection on the smoothed gray level image by using a Canny edge detection algorithm to obtain a black spot area edge image;
Counting the number of all pixels in the obtained black spot area edge image, calculating the standard deviation of pixel gradient values of all pixels, comparing the calculated standard deviation with a standard deviation threshold value, and judging that the type of the black spot area is the black spot belonging to the chilli per se if the standard deviation of the pixel grid is higher than the standard deviation threshold value, otherwise, judging that the type of the black spot area is the other type.
In step 103, a method of constructing a black spot classification model, comprising the steps of:
step 201, constructing pepper individual appearance image map structure data;
Dividing the pepper individual appearance image into a plurality of small areas;
Each small block is used as a node in the graph, the small block nodes adjacent to each other are connected according to the spatial position relation between the small block areas as edges, and the pepper individual appearance image graph structure is constructed;
Step 202, a black spot classification model comprises the following steps: the method comprises the steps of inputting a node characteristic matrix and an adjacent matrix into a picture scroll lamination layer, outputting a new node characteristic matrix by the picture scroll lamination layer, inputting the new node characteristic matrix into the picture scroll lamination layer, outputting a node characteristic matrix after downsampling by the picture scroll lamination layer, inputting the node characteristic matrix after downsampling into the full connection layer, outputting two classifications by the full connection layer, and respectively corresponding to 'pixels belonging to adhered impurities' and 'black spots belonging to peppers' in 'black spot' types in individual appearance images of peppers by two classification labels;
And 203, constructing a training data set and training a black spot classification model.
In step 102, the pepper appearance image is a top view image of the pepper on the production line.
The calculation formula of the graph convolution layer is as follows:
wherein, An index representing the current layer is shown,Represents a set of nodes adjacent to node v,Representing adjacent nodesThe root number of the product of the degrees,A trainable weight matrix representing the current layer,Representing the trainable bias vector for the current layer,Representing the ReLU function.
The pooling layer adopts a flat pooling mode, and directly splices all node representations to be used as a new graph representation; or in a hierarchical pooling manner, centering on a node representation, node features adjacent to the node representation are aggregated/pooled into a new node representation through a max operator.
The calculation formula of the full connection layer is as follows:
wherein, Representing the input downsampled node feature matrix,Representing a matrix of trainable weights,Representing the trainable bias vector and,As a softmax function.
A monitoring management system for pepper food production and processing comprises the following modules:
the data acquisition module is used for acquiring pepper information;
The image marking module is used for marking the black spot areas in all the pepper individual appearance images at the pixel level;
the black spot classification generation module is used for generating a black spot classification model;
and the execution module is used for selecting a processing mode of the corresponding pepper individual according to the classification result of the pepper individual appearance image output by the black spot classification model.
A storage medium storing non-transitory computer-readable instructions that, when executed by a computer, are capable of performing steps in a monitoring and management method for pepper food production processing as described above.
The invention has the beneficial effects that: the monitoring and managing method can accurately identify the pepper individuals with impurities on the surfaces and still can be used, and avoids the situation that the pepper individuals are mistakenly removed together with peppers with black spots formed by decay, thereby reducing the waste of raw materials and improving the utilization rate of the raw materials.
Drawings
Fig. 1 is a flowchart of the monitoring and managing method for producing and processing chilli food according to the present invention.
FIG. 2 is a flow chart of a method of constructing a black spot classification model in accordance with the present invention.
Fig. 3 is a block diagram of the monitoring and management system for producing and processing chilli food according to the present invention.
In the figure: 100. a data acquisition module; 200. an image marking module; 300. a black spot classification generation module; 400. and executing the module.
Detailed Description
The subject matter described herein will now be discussed with reference to example embodiments. It is to be understood that these embodiments are merely discussed so that those skilled in the art may better understand and implement the subject matter described herein and that changes may be made in the function and arrangement of the elements discussed without departing from the scope of the disclosure herein. Various examples may omit, replace, or add various procedures or components as desired. In addition, features described with respect to some examples may be combined in other examples as well.
As shown in fig. 1, the monitoring and managing method for producing and processing the chilli food comprises the following steps:
Step 101, acquiring pepper information, wherein the pepper information comprises pepper variety information, growing environment information and pepper appearance images;
The variety of the peppers is the same as that of the same batch of peppers, and the variety information of the peppers is expressed in a text form;
The information of the growing environment of the peppers comprises impurity types existing in the growing environment of the peppers, the impurity types comprise sediment, dead branches, fallen leaves and the like, and when the peppers are planted in a greenhouse, the impurity types are mainly stems and leaves of the peppers; when the capsicum is planted outdoors, the impurity type is mainly sediment;
the appearance image information of the pepper comprises color information of black spots on the surface of the pepper, shape and size information of the black spots on the surface of the pepper and distribution rule information of the black spots on the surface of the pepper;
In one embodiment of the invention, the pepper appearance image is a top view image of the peppers on a production line, in order to avoid overlapping of a plurality of peppers, each pepper can be placed at intervals, only one pepper in the pepper appearance image obtained by photographing each time is ensured, the placement mode of the peppers can be not limited, and the appearance image of a single pepper individual can be obtained from the pepper appearance image through image segmentation.
Acquiring color information of black spots on the surface of the pepper, shape and size information of the black spots on the surface of the pepper and distribution rule information of the black spots on the surface of the pepper from the pepper appearance image;
102, performing image segmentation on the acquired pepper appearance images to obtain all pepper individual appearance images, performing pixel-level labeling on black spot areas in all the pepper individual appearance images, and marking each pixel of the black spot areas in the pepper individual appearance images as one of black spots belonging to peppers or adhering impurities;
It should be noted that, firstly, the types of marked pixels of the black spot area in the appearance image of the pepper individual are summarized, if the number of the marked pixels of the black spot area is more than half of the number of the types of the black spots belonging to the pepper, the black spot area is marked as the black spot belonging to the pepper, otherwise, the black spot area is marked as the adhered impurity;
the rule of the mark is: black spots generated by mildew of pepper individuals tend to be irregular in shape and blurred in edges, and impurities are adhered to the pepper individuals generally in regular geometric shapes, so that the edges are clear.
For example, 1 pepper individual appearance image is obtained by collecting pepper appearance images and performing image division, and there is a "black spot area" in the pepper individual appearance image, the number of pixels of the black spot area of the "black spot area" is 17 by edge detection and contour analysis, wherein the pixels of 14 black spot areas are marked as pixels belonging to the black spot of the pepper itself, the pixels of 3 black spot areas are marked as pixels belonging to the adhered impurities, and the category of the "black spot area" needs to be marked as black spot belonging to the pepper itself.
In one embodiment of the present invention, a method for processing the collected pepper appearance image in step 102 to obtain a black spot area category is provided, which comprises the following steps:
1. Manually labeling a seed dataset;
Firstly, manually marking the type of black spots (the black spots belonging to the peppers or the adhered impurities) in a part of pepper images to be used as a seed training set with labels;
For example, several workers were asked to mark the black spot type in 500 pepper raw material images as a labeled seed training set.
2. Training an initial classifier;
Training a ResNet-based convolutional neural network model by using a seed training set to serve as an initial model of black spot classification;
predicting unlabeled data, and performing black spot type prediction on a large number of unlabeled pepper images by using a trained initial model to obtain pseudo tags; performing black spot type prediction on 10 ten thousand unlabeled pepper images acquired in 1 month of a factory by using an initial model, and generating a pseudo tag for each sample;
3. Constructing a training set;
combining a sample with higher prediction confidence (such as credibility TOP 30%) and a pseudo tag thereof with an original seed subset to form an expanded semi-supervised training set;
3 ten thousand pseudo tag samples with 30% confidence coefficient TOP are selected according to the sequence of the model prediction confidence coefficient from high to low, and are combined with the original seed set to construct a semi-supervised training set;
4. Training a semi-supervised model;
Based on the expanded semi-supervised training set, the model is subjected to semi-supervised training on the expanded training set by using an inconsistent regularization strategy of a semi-supervised learning algorithm, and a final black spot classifier is obtained.
In another embodiment of the present invention, there is further provided a method for processing the collected appearance image of the pepper individual in step 102 to obtain a black spot area category, including the following steps:
Step one, carrying out gray conversion on a large number of collected pepper appearance images, and converting color images into gray images;
Secondly, performing edge detection on the smoothed gray level image by using a Canny edge detection algorithm to obtain a black spot area edge image;
Counting the number of all pixels in the obtained black spot area edge image, calculating the standard deviation of pixel gradient values of all pixels by adopting a Gaussian gradient method, comparing the standard deviation obtained by calculation with a standard deviation threshold value, and judging that the type of the black spot area is the black spot belonging to the chilli per se if the standard deviation of the pixel grid is higher than the standard deviation threshold value;
Specifically, as black spots generated by mildew of the capsicum are irregular in shape and blurred in edge, and impurities are adhered to the capsicum, the phenomenon is that the edges are clear, in one embodiment of the invention, the method comprises the steps of calculating the pixel gradient value in the neighborhood around each pixel grid in the black spot area edge image in the second step, marking the pixel grid in the center of the capsicum based on the size result between the standard deviation and the standard deviation threshold value obtained by comparison calculation, counting the proportion of the number of the pixel grids of the same type to the number of all pixels in the obtained black spot area edge image, and judging that the black spot area in the capsicum appearance image is of the type if the proportion is more than 50%, otherwise, judging that the black spot area in the capsicum appearance image is of the other type;
the following provides a method for calculating the pixel gradient value of each pixel grid in the black spot region edge image:
1. selecting a neighborhood window with proper size, for example 3*3, and performing sobel gradient calculation on the pixel values in the window to obtain the gradient value of the pixel grid;
2. Calculating standard deviation for gradient values of all pixels in the window;
3. Setting a standard deviation threshold, judging that irregular blurred edges are black spots generated by mildew if the standard deviation or entropy of a certain pixel grid is higher than the threshold, and judging that the regular geometric clear edges formed by adhering impurities are black spots if the standard deviation or entropy of the certain pixel grid is lower than the threshold.
For example: given a pepper appearance image, an analysis was performed on a small 10 x 10 area containing a mildew-generated irregular blur "black spot" with pixel values as follows (0 for black and 255 for white):
| 248 | 250 | 252 | 248 | 246 | 242 | 235 | 230 | 228 | 226 |
| 251 | 253 | 255 | 249 | 244 | 237 | 228 | 222 | 219 | 217 |
| 254 | 256 | 258 | 252 | 247 | 239 | 231 | 224 | 220 | 215 |
| 255 | 257 | 259 | 253 | 248 | 241 | 234 | 226 | 219 | 213 |
| 256 | 258 | 260 | 254 | 250 | 243 | 236 | 229 | 221 | 214 |
| 257 | 259 | 261 | 255 | 251 | 245 | 238 | 231 | 223 | 216 |
| 258 | 260 | 262 | 256 | 252 | 247 | 240 | 233 | 225 | 218 |
| 259 | 261 | 263 | 257 | 253 | 248 | 242 | 235 | 227 | 220 |
| 260 | 262 | 264 | 258 | 254 | 249 | 243 | 237 | 229 | 222 |
| 261 | 263 | 265 | 259 | 255 | 250 | 244 | 238 | 231 | 224 |
analyzing the pixel values of the areas, the pixel values in the black spot areas are lower, and the edge transition is fuzzy;
Then, calculating the standard deviation of pixel gradient values in a 3*3 neighborhood window of each pixel grid in the region to obtain the following data:
| 12.5 | 14.2 | 16.7 | 19.5 | 22.6 | 24.8 | 25.3 | 24.1 | 21.9 | 18.4 |
| 15.8 | 17.3 | 19.2 | 21.5 | 23.8 | 25.1 | 24.8 | 23.2 | 20.7 | 17.4 |
| 18.6 | 19.6 | 21.0 | 22.7 | 24.3 | 24.9 | 24.3 | 22.6 | 20.2 | 17.2 |
| 20.5 | 21.1 | 22.1 | 23.3 | 24.4 | 24.6 | 23.9 | 22.3 | 20.1 | 17.5 |
| 21.6 | 22.0 | 22.7 | 23.5 | 24.2 | 24.2 | 23.6 | 22.2 | 20.3 | 18.1 |
| 21.9 | 22.3 | 22.9 | 23.4 | 23.8 | 23.8 | 23.3 | 22.2 | 20.6 | 18.7 |
| 21.6 | 22.0 | 22.6 | 23.0 | 23.2 | 23.2 | 22.9 | 22.1 | 20.8 | 19.2 |
| 20.7 | 21.2 | 21.9 | 22.3 | 22.5 | 22.5 | 22.3 | 21.8 | 20.9 | 19.6 |
| 19.3 | 20.0 | 20.8 | 21.4 | 21.7 | 21.8 | 21.7 | 21.4 | 20.8 | 19.8 |
| 17.4 | 18.3 | 19.3 | 20.1 | 20.6 | 20.9 | 21.0 | 20.8 | 20.4 | 19.7 |
setting the standard deviation threshold value as 22, wherein the standard deviation of the pixel grids in the black spot area is generally higher than 22.0, and the standard deviation of the boundary pixel grids is also mostly higher than the threshold value; therefore, most of the pixel cells in the "black spot" region of this irregularly blurred edge are determined to be mildew-generated irregularly blurred "black spots".
In one embodiment of the invention, a dataset is established according to the annotated pepper individual appearance image, and the dataset is divided into a training set, a verification set and a test set.
Step 103, as shown in fig. 2, modeling is performed based on the marked pepper individual appearance image, and a black spot classification model is generated;
The method for constructing the black spot classification model comprises the following steps:
step 201, constructing pepper individual appearance image map structure data;
Dividing the pepper individual appearance image into a plurality of small areas;
Each small block is used as a node in the graph, the small block nodes adjacent to each other are connected according to the spatial position relation between the small block areas as edges, and the pepper individual appearance image graph structure is constructed;
For the first A pixel block, whereinThe pixel block represents the pixel block of the ith row and the jth column, creating a nodeThen nodeIs: ; wherein, Representing nodesIs characterized by the mean value of (c),Representing nodesIs characterized by the standard deviation of (c) in terms of,、、Representing nodes respectivelyFeatures in R, G, B in the RGB channel;
For each node With nodes8 Adjacent nodes on the image, noted as; If it isIs thatIs in the vicinity of the node(s)AndAn undirected edge is added between the two。
For example, the pepper individual appearance image is a 512x512 pixel RGB image, the image contains a "black spot" area, the image is divided into 64 (8×8) non-overlapping 64×64 pixel blocks, and each pixel block is regarded as a node in the image structure; nodes between adjacent pixel blocks are connected to form the structure data of the pepper individual appearance image map; then through the steps, an undirected graph with 64 nodes and 256 edges is constructedWherein, the method comprises the steps of, wherein,Representing a constructed undirected graph with 64 nodes and 256 edges,Representing a set of all the nodes,Representing a set of all edges.
Step 202, a black spot classification model comprises the following steps: the method comprises the steps of inputting a node characteristic matrix and an adjacent matrix into a picture rolling layer, outputting a new node characteristic matrix by the picture rolling layer, inputting the new node characteristic matrix into the picture rolling layer, outputting a down-sampled node characteristic matrix by the picture rolling layer, inputting the down-sampled node characteristic matrix into the full connecting layer, outputting two classifications by the full connecting layer, and respectively corresponding to ' pixels belonging to adhered impurities ' and ' black spots belonging to peppers ' in ' black spot ' types in individual appearance images of peppers ' by two classification labels represented by the two classifications;
The calculation formula of the graph convolution layer is as follows:
wherein, Represent the firstLayer nodeIs provided with a hidden feature of (a),An index representing the current layer is shown,Represents a set of nodes adjacent to node v,Representing a node adjacent to the node v,Representing adjacent nodesThe root number of the product of the degrees,A trainable weight matrix representing the current layer,Node representing current layerIs provided with a hidden feature of (a),Representing the trainable bias vector for the current layer,Representing a ReLU function;
The pooling layer adopts a flat pooling mode, and directly splices all node representations to be used as a new graph representation; or in a hierarchical pooling manner, centering on a node representation, node features adjacent to the node representation are aggregated/pooled into a new node representation through a max operator.
The calculation formula of the full connection layer is as follows:
wherein, A classification value representing the input individual appearance image of the capsicum,Representing the input downsampled node feature matrix,Representing a matrix of trainable weights,Representing the trainable bias vector and,As a softmax function;
step 203, constructing a training data set and training a black spot classification model;
Collecting a large number of pepper individual appearance images containing various black spots, and constructing a graph structure according to step 201 for each image to obtain node characteristics and node connection relations; labeling corresponding class labels for the black spot condition of each image, wherein the class labels comprise pixels belonging to adhered impurities and black spots belonging to peppers, inputting constructed graph structure data into a black spot classification model, setting a loss function (such as cross entropy loss) by using labels supervision, selecting an Adam optimizer, setting super-parameters such as learning rate and the like, performing iterative training in a batch mode, minimizing training loss, evaluating the model on a verification set regularly, and storing the model with the best performance on the verification set;
For a new pepper image, constructing a graph structure and node characteristics according to step 201;
the graph data are input into a trained GCN model, and the output new pepper image is about the prediction result of the 'black spot' type.
Step 104, deploying the black spot classification model into an automatic detection system of a production line, and selecting a corresponding processing mode of the pepper individuals by using classification results of the pepper individual appearance images output by the black spot classification model;
Specifically, according to the prediction result of the black spot classification model, classifying the capsicum individuals corresponding to the capsicum image, directly removing and cleaning the capsicum classified as the black spot belonging to the capsicum, and sending the capsicum classified as the adhered impurity into a further cleaning link;
it should be noted that, the individual capsicum without "black spot" in the capsicum image is removed on the production line before entering the black spot classification model for classification.
As shown in fig. 3, a monitoring and management system for producing and processing chilli food comprises the following modules:
the data acquisition module 100 is used for acquiring pepper information;
the image marking module 200 is used for marking the black spot areas in all the pepper individual appearance images at the pixel level;
the black spot classification generation module 300 is used for generating a black spot classification model;
The execution module 400 is configured to select a processing mode of the corresponding pepper individual according to the classification result of the pepper individual appearance image output by the black spot classification model.
A storage medium storing non-transitory computer-readable instructions that, when executed by a computer, are capable of performing steps in a monitoring and management method for pepper food production processing as described above.
The embodiment has been described above with reference to the embodiment, but the embodiment is not limited to the above-described specific implementation, which is only illustrative and not restrictive, and many forms can be made by those of ordinary skill in the art, given the benefit of this disclosure, are within the scope of this embodiment.
Claims (8)
1. The monitoring and managing method for the production and processing of the chilli food is characterized by comprising the following steps of:
Step 101, acquiring pepper information, wherein the pepper information comprises pepper variety information, growing environment information and pepper appearance images;
102, performing image segmentation on the acquired pepper appearance images to obtain all pepper individual appearance images, performing pixel-level labeling on black spot areas in all the pepper individual appearance images, marking each pixel of the black spot areas in the pepper individual appearance images as black spots belonging to peppers or one of two adhered impurities, and performing pixel-level labeling on the black spot areas in all the pepper individual appearance images, wherein the labeling rule is that firstly, the marked types of the pixels of the black spot areas in the pepper individual appearance images are summarized, if the number of the marked pixels of the black spot areas is more than half of the number of the marked types of the black spots belonging to peppers, the black spot areas are marked as black spots belonging to peppers, otherwise, the black spot areas are marked as adhered impurities;
step 103, modeling based on the marked pepper individual appearance image to generate a black spot classification model;
in step 103, a method of constructing a black spot classification model, comprising the steps of:
step 201, constructing pepper individual appearance image map structure data;
Dividing the pepper individual appearance image into a plurality of small areas;
Each small block is used as a node in the graph, the small block nodes adjacent to each other are connected according to the spatial position relation between the small block areas as edges, and the pepper individual appearance image graph structure is constructed;
Step 202, a black spot classification model comprises the following steps: the method comprises the steps of inputting a node characteristic matrix and an adjacent matrix into a picture rolling layer, outputting a new node characteristic matrix by the picture rolling layer, inputting the new node characteristic matrix into the picture rolling layer, outputting a node characteristic matrix after downsampling by the picture rolling layer, inputting the node characteristic matrix after downsampling into the full connecting layer, outputting two classifications by the full connecting layer, wherein two classification labels expressed by the two classifications correspond to black spot types of black spot areas of pepper individuals in pepper individual appearance images respectively, and the black spot types of the black spot areas comprise adhered impurities and black spots belonging to the peppers;
step 203, constructing a training data set and training a black spot classification model;
and 104, deploying the black spot classification model into an automatic detection system of a production line, and selecting a corresponding processing mode of the pepper individuals by using the classification result of the pepper individual appearance image output by the black spot classification model.
2. The method for monitoring and managing the processing of pepper foods as recited in claim 1, wherein in the step 102, the method for obtaining the black spot area category by processing the appearance image of the individual peppers further comprises the following steps:
step one, carrying out gray conversion on the collected pepper appearance image, and converting a color image into a gray image;
Secondly, performing edge detection on the smoothed gray level image by using a Canny edge detection algorithm to obtain a black spot area edge image;
counting the number of all pixels in the obtained black spot area edge image, calculating the standard deviation of pixel gradient values of all pixels by adopting a Gaussian gradient method, comparing the calculated standard deviation with a standard deviation threshold value, and judging that the type of the black spot area is marked as the black spot belonging to the pepper per se if the standard deviation of the pixel grid is higher than the standard deviation threshold value, otherwise, judging that the type of the black spot area is the other type.
3. A method for monitoring and controlling the production and processing of pepper foods as claimed in claim 1 or 2, wherein in step 102, the appearance image of the pepper is a top view image of the pepper on the production line.
4. A method for monitoring and managing the processing of pepper food as claimed in claim 3, characterized in that the calculation formula of the graph convolution layer is:
；
wherein, Represent the firstLayer nodeIs provided with a hidden feature of (a),An index representing the current layer is shown,Represents a set of nodes adjacent to node v,Representing a node adjacent to the node v,Representing adjacent nodesThe root number of the product of the degrees,A trainable weight matrix representing the current layer,Node representing current layerIs provided with a hidden feature of (a),Representing the trainable bias vector for the current layer,Representing the ReLU function.
5. The method for monitoring and managing the production and processing of pepper foods as claimed in claim 4, wherein the pooling layer adopts a flat pooling mode, and all node representations are directly spliced together to be used as new map representations; or in a hierarchical pooling manner, centering on a node representation, and node features adjacent to the node representation are aggregated or pooled into a new node representation through a max operator.
6. The method for monitoring and managing the production and processing of pepper foods as claimed in claim 5, wherein the calculation formula of the fully-connected layer is as follows:
；
wherein, A classification value representing the input individual appearance image of the capsicum,Representing the input downsampled node feature matrix,Representing a matrix of trainable weights,Representing the trainable bias vector and,As a softmax function.
7. A monitoring and management system for producing and processing a capsicum food, which adopts the monitoring and management method for producing and processing a capsicum food as claimed in claim 6, comprising the following modules:
the data acquisition module (100) is used for acquiring pepper information;
the image marking module (200) is used for marking the black spot areas in all the pepper individual appearance images at the pixel level;
A black spot classification generation module (300) for generating a black spot classification model;
And the execution module (400) is used for selecting a corresponding processing mode of the pepper individual according to the classification result of the pepper individual appearance image output by the black spot classification model.
8. A storage medium storing non-transitory computer-readable instructions, wherein the non-transitory computer-readable instructions, when executed by a computer, are capable of performing a monitoring and management method for pepper food production as recited in claim 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410544877.0A CN118135339B (en) | 2024-05-06 | Monitoring management method and system for chilli food production and processing |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410544877.0A CN118135339B (en) | 2024-05-06 | Monitoring management method and system for chilli food production and processing |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN118135339A CN118135339A (en) | 2024-06-04 |
| CN118135339B true CN118135339B (en) | 2024-07-09 |
Family
ID=
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112036335A (en) * | 2020-09-03 | 2020-12-04 | 南京农业大学 | Deconvolution-guided semi-supervised plant leaf disease identification and segmentation method |
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112036335A (en) * | 2020-09-03 | 2020-12-04 | 南京农业大学 | Deconvolution-guided semi-supervised plant leaf disease identification and segmentation method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112036335B (en) | Inverse convolution guided semi-supervised plant leaf disease identification and segmentation method | |
| CN111768432A (en) | Moving target segmentation method and system based on twin deep neural network | |
| CN109766898A (en) | Image character recognition method, device, computer equipment and storage medium | |
| CN106340016A (en) | DNA quantitative analysis method based on cell microscope image | |
| CN110503140B (en) | Deep migration learning and neighborhood noise reduction based classification method | |
| CN111986125A (en) | Method for multi-target task instance segmentation | |
| CN112419202A (en) | Wild animal image automatic identification system based on big data and deep learning | |
| CN111241933A (en) | Pig farm target identification method based on universal countermeasure disturbance | |
| CN114241425B (en) | Training method and device of garbage detection model, storage medium and equipment | |
| CN116129426A (en) | Fine granularity classification method for cervical cell smear 18 category | |
| CN113159159B (en) | Small sample image classification method based on improved CNN | |
| CN118135339B (en) | Monitoring management method and system for chilli food production and processing | |
| CN110349119B (en) | Pavement disease detection method and device based on edge detection neural network | |
| CN118135339A (en) | Monitoring management method and system for chilli food production and processing | |
| CN115641575A (en) | Leafhopper agricultural pest detection method based on sparse candidate frame | |
| CN112785559B (en) | Bone age prediction method based on deep learning and formed by mutually combining multiple heterogeneous models | |
| CN114937030A (en) | Phenotypic parameter calculation method for intelligent agricultural planting of lettuce | |
| CN114511849A (en) | Grape thinning identification method based on graph attention network | |
| Fan et al. | Recognition and analysis of fabric texture by double-sided fusion of transmission and reflection images under compound light source | |
| CN116188774B (en) | Hyperspectral image instance segmentation method and building instance segmentation method | |
| CN117218452B (en) | Automatic classification management system for land images | |
| CN111242895B (en) | Bamboo chip wormhole and mildew detection method based on convolution flexible neural forest | |
| Monteiro et al. | Deep learning recognition of a large number of pollen grain types | |
| CN111709933B (en) | Method for detecting ream paper defect segmentation classification | |
| CN111353966B (en) | Image fusion method based on total variation deep learning and application and system thereof |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant |