CN111507379A - Ore automatic identification and rough sorting system based on deep learning - Google Patents
Ore automatic identification and rough sorting system based on deep learning Download PDFInfo
- Publication number
- CN111507379A CN111507379A CN202010215037.1A CN202010215037A CN111507379A CN 111507379 A CN111507379 A CN 111507379A CN 202010215037 A CN202010215037 A CN 202010215037A CN 111507379 A CN111507379 A CN 111507379A
- Authority
- CN
- China
- Prior art keywords
- ore
- mud
- deep learning
- model
- training
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B07—SEPARATING SOLIDS FROM SOLIDS; SORTING
- B07C—POSTAL SORTING; SORTING INDIVIDUAL ARTICLES, OR BULK MATERIAL FIT TO BE SORTED PIECE-MEAL, e.g. BY PICKING
- B07C5/00—Sorting according to a characteristic or feature of the articles or material being sorted, e.g. by control effected by devices which detect or measure such characteristic or feature; Sorting by manually actuated devices, e.g. switches
- B07C5/36—Sorting apparatus characterised by the means used for distribution
- B07C5/361—Processing or control devices therefor, e.g. escort memory
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B07—SEPARATING SOLIDS FROM SOLIDS; SORTING
- B07C—POSTAL SORTING; SORTING INDIVIDUAL ARTICLES, OR BULK MATERIAL FIT TO BE SORTED PIECE-MEAL, e.g. BY PICKING
- B07C5/00—Sorting according to a characteristic or feature of the articles or material being sorted, e.g. by control effected by devices which detect or measure such characteristic or feature; Sorting by manually actuated devices, e.g. switches
- B07C5/36—Sorting apparatus characterised by the means used for distribution
- B07C5/361—Processing or control devices therefor, e.g. escort memory
- B07C5/362—Separating or distributor mechanisms
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computational Linguistics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Evolutionary Biology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses an ore automatic identification and rough sorting system based on deep learning, which comprises: the model building module is used for building a deep learning model for automatic ore identification and rough sorting; the model training module is used for forming a training set and a testing set according to the selected picture of the ore mud block mixture and training a deep learning model for automatic ore recognition and rough sorting; the recognition module is used for inputting the shot ore-mud block mixture picture on the crawler belt into a trained deep learning model for automatic ore recognition and coarse sorting to obtain a real-time recognition result; the sorting module is used for controlling the crawler belt to send the ore of the batch into the next procedure according to the recognition result of the recognition module if the mud content after recognition is smaller than a set threshold value, otherwise, controlling the high-pressure water gun to carry out fixed-point flushing on the recognized mud blocks and then sending the mud blocks to the next procedure. The invention creatively introduces deep learning into the ore sorting process, and realizes the full automation of ore coarse sorting.
Description
Technical Field
The invention relates to an ore sorting technology, in particular to an ore automatic identification and rough sorting system based on deep learning.
Background
After ore mining, the first procedure is to crush the ore and mud blocks, after crushing, the ore washing needs to be carried out for several times, the first two to three times of ore washing rough separation are carried out, and a high-pressure water gun is used for washing loose mud. Moreover, in the process of ore rough separation, the problems of overlarge water resource consumption, low automation and intelligence degree caused by more manual participation and the like also exist. How to solve the above problems by using the current new technology is the matter that the invention is to solve.
Disclosure of Invention
The invention aims to solve the technical problem of providing an automatic ore identification and rough sorting system based on deep learning aiming at the defects in the prior art.
The technical scheme adopted by the invention for solving the technical problems is as follows: an automatic ore identification and coarse sorting system based on deep learning, comprising:
the model building module is used for building a deep learning model for automatic ore identification and rough sorting; the deep learning model for ore automatic identification and rough sorting comprises a classification model network and a target detection framework;
the classification model network is used for extracting deep layer network features and shallow layer network features of ores and mud blocks in the training image;
the target detection framework is used for predicting and outputting a mud block identification result of the ore and mud block mixture according to the deep layer network characteristics and the shallow layer network characteristics of the extracted ore and mud blocks;
the model training module is used for forming a training set and a testing set according to the selected picture of the ore mud block mixture and training a deep learning model for automatic ore recognition and rough sorting;
the recognition module is used for inputting the shot ore-mud block mixture picture on the crawler belt into a trained deep learning model for automatic ore recognition and coarse sorting to obtain a real-time recognition result;
the sorting module is used for sending the ore of the batch into the next procedure if the mud content is less than 3% of the threshold value after the ore is identified according to the identification result of the identification module, otherwise, controlling the high-pressure water gun to carry out fixed-point flushing on the identified mud block and then sending the mud block to the next procedure.
According to the scheme, the classification model network in the step 1) is a VGG16 network and is used for extracting deep layer network features and shallow layer network features of ores and mud blocks in the training image; and the target detection framework is an SSD512 training model and is used for predicting and outputting the recognition results of the ores and the mud blocks according to the deep layer network characteristics and the shallow layer network characteristics of the extracted ores and mud blocks.
According to the scheme, the pictures of the ore mud block mixture are selected from the model training module to form a training set and a testing set, and the deep learning model of ore automatic identification and rough sorting is trained, and the method specifically comprises the following steps:
1) marking the collected ore and mud block mixture data set according to the data set format adopted by the SSD official model to obtain the target marked 4-tuple parameter (X)min，Ymin，Xmax，Ymax) Respectively representing the coordinates of the upper left corner and the lower right corner of the marking frame, namely only two Classes of Mud blocks Mud and Background are available, and the information is stored in an xml file;
2) the training set and the test set are as follows: 2, dividing in proportion, wherein a training set is used for training network parameters, and a test set is used for performance evaluation of target prediction;
3) inputting the data set into a target detection framework, after image data is input into a network, continuously learning the network, continuously updating parameters, continuously improving the regression of positions and the classification accuracy of categories through the feedforward of the network, combining a loss function and an SGD gradient descent method, and finally obtaining an ideal model weight.
According to the scheme, the data set in the step 1) is expanded in the following mode: using a transformation function, flipping, cropping, or panning the marked image results in added data.
According to the scheme, the image data input network in the step 3) can carry out random image enhancement transformation to obtain increased image data, and the data volume of the data set is expanded.
The invention also provides an ore automatic identification and rough sorting method based on deep learning, which is characterized by comprising the following steps:
1) constructing a deep learning model for automatic ore identification and rough sorting; the deep learning model for ore automatic identification and rough sorting comprises a classification model network and a target detection framework;
the classification model network is used for extracting deep layer network features and shallow layer network features of ores and mud blocks in the training image;
the target detection framework is used for predicting and outputting a mud block identification result of the ore and mud block mixture according to the deep layer network characteristics and the shallow layer network characteristics of the extracted ore and mud blocks;
2) selecting pictures of the ore mud block mixture to form a training set and a testing set, and training a deep learning model for automatic ore identification and rough sorting;
3) the method comprises the steps of shooting an ore mud block mixture picture on a crawler belt, inputting the picture into a trained deep learning model for automatic ore identification and coarse sorting, and obtaining a real-time identification result;
4) according to the identification result, if the mud content after identification is less than the set threshold value, the crawler belt is controlled to send the ore of the batch into the next procedure, otherwise, the high-pressure water gun is controlled to carry out fixed-point flushing on the identified mud blocks, and then the mud blocks are sent to the next procedure.
According to the scheme, the classification model network in the step 1) is a VGG16 network and is used for extracting deep layer network features and shallow layer network features of ores and mud blocks in the training image; and the target detection framework is an SSD512 training model and is used for predicting and outputting the recognition results of the ores and the mud blocks according to the deep layer network characteristics and the shallow layer network characteristics of the extracted ores and mud blocks.
According to the scheme, the pictures of the ore mud block mixture are selected from the model training module to form a training set and a testing set, and the deep learning model of ore automatic identification and rough sorting is trained, and the method specifically comprises the following steps:
2.1) marking the collected ore and mud block mixture data set according to the data set format adopted by the SSD official model to obtain the target marked 4-tuple parameter (X)min，Ymin，Xmax，Ymax) Respectively representing the coordinates of the upper left corner and the lower right corner of the marking frame, namely only two Classes of Mud blocks Mud and Background are available, and the information is stored in an xml file;
2.2) the training set and the test set are adjusted according to the weight ratio of 8: 2, dividing in proportion, wherein a training set is used for training network parameters, and a test set is used for performance evaluation of target prediction;
and 2.3) inputting the data set into a target detection framework, after image data are input into a network, continuously learning the network, continuously updating parameters, continuously improving the regression of positions and the classification accuracy of categories through the feedforward of the network, combining with a loss function and an SGD gradient descent method, and finally obtaining an ideal model weight.
According to the scheme, the step 2.1) further comprises the step of expanding the collected ore-mud block mixture data set, and the data set is expanded in the following mode: using a transformation function, flipping, cropping, or panning the marked image results in added data.
According to the scheme, the image data input network in the step 2.3) carries out random image enhancement transformation to obtain increased image data, and the data volume of the data set is expanded.
The invention has the following beneficial effects: by using the ore automatic identification and rough sorting system based on deep learning, the manual participation degree in the ore rough sorting process is greatly reduced, and the labor intensity of related workers is reduced. In addition, the invention creatively introduces deep learning into the ore sorting process, realizes full automation of ore coarse sorting, can realize real-time monitoring operation of more than 30fps through a deep learning model, realizes fixed-point quantitative use of resources such as water, electricity and the like on the basis of a detection result, saves sorting cost and improves environmental benefits.
Drawings
The invention will be further described with reference to the accompanying drawings and examples, in which:
FIG. 1 is a flow chart of a method of an embodiment of the present invention;
FIG. 2 is a schematic structural diagram of an embodiment of the present invention;
FIG. 3 is a diagram of a classification model network and object detection framework provided by the present invention;
fig. 4 is a schematic diagram of the operation of the ore automatic identification and rough sorting system based on deep learning provided by the invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail with reference to the following embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
As shown in fig. 1 and 4, an automatic ore identification and rough sorting method based on deep learning includes the following steps:
s1, washing the ore mud block mixture just sent by using a high-pressure water gun, wherein the washing time is generally controlled to be 3 minutes;
s2, the lower computer controls the camera module to take pictures of the ore on the track and transmits the pictures to the upper computer through a network;
s3, storing the pictures in a database by the upper computer, identifying the pictures by using a trained deep learning model, and measuring and calculating the soil content of the ore in the batch;
s4, if the calculated soil content is less than the threshold value (generally set as 3%), entering a step S5, otherwise, entering a step S6;
s5, the upper computer sends an appointed control signal to the lower computer to control the crawler belt, the ore of the batch is conveyed to the next procedure, and the ore washing and roughing of the batch are completed;
and S6, after the upper computer identifies the mud blocks, sending information containing the positions of the mud blocks to the lower computer, controlling a high-pressure water gun by the lower computer, washing the mud blocks at the specified positions in a fixed-point quantitative mode, and then carrying out steps S2 to S4 after washing.
In step S2, before the system starts working, the system is initialized, after the system is initialized, an OV2640 module is used to collect the image of the ore mud block, 1920 × 1080 resolution is adopted, after the collection starts, whether the collection of the image of one frame is completed is judged, after the collection is completed, a single chip (i.e. a lower computer) can process the image, since the data of one frame of image far exceeds the limit of the maximum number 65535 bytes of DMA transmission, a plurality of times of DMA interruptions are required to be completed, the image data can be stored in an SRAM only by changing the storage address of the image data during the interruption, uncompressed RGB565 image format data flow is too large, which can hinder the transmission and storage of the image, the device compresses the image in JPEG format, the original image is subjected to image preprocessing, discrete cosine change, quantization and quantization, and finally encapsulates the compressed data into a JPEG file header, a file tail, a data segment, i.e. an entropy coding is a standard JPEG format file, then the WiFi module is sent out, the WiFi module supports 802.11b/g/n Socket, the wireless communication is installed, a wireless network header, a TCP data segment is installed, a wireless network is displayed as a wireless network, a wireless network communication protocol is displayed, and a wireless access point is displayed, a wireless network is displayed, and a wireless network is displayed clearly displayed, a wireless access point is displayed, a wireless network is displayed, a.
In step S3, as shown in fig. 3, the deep learning model uses VGG16 classification network and SSD512 target detection framework. The VGG16 is a classical convolutional neural network and is composed of an input layer, a convolutional layer, a full connection layer and an output layer, wherein the input layer is used for directly inputting original data; the convolutional layer is mainly used for extracting input data features and comprises two parts, wherein one part is a convolutional core, and the other part is a downsampling layer, also called Pooling layer, and is used for reducing the dimension of a target image so as to reduce the data processing amount; the full connection layer is equivalent to a classifier and is used for realizing the longitudinal conduction of signals, the neuron nodes of each layer are respectively connected with the weights on the lines, and then the weighted combination is carried out to obtain the input of the neuron nodes of the next layer; the output layer is used for outputting the result. The SSD512 model enables the network to extract more subtle features and expands the number of prior frames, so that the identification accuracy of the model is improved by one level. And in the process of model training, testing the accuracy of the model at intervals. During testing, the test set picture is preprocessed and input into the trained SSD512 network model, and an output prediction frame position loc value and a confidence conf value are obtained. The loc value is decoded (decode) and compared with 24562 a priori boxes, and boxes with conf less than a given threshold are screened out by a screening rule. A number of overlapping boxes are available and the final boxed result can be obtained by applying an NMS (non-maximum suppression) algorithm to these boxes. And analyzing the learning condition of the network at the moment according to the prediction result.
NMS algorithm flow: assuming that there are N frames, the classification score of each frame is Si (1 ═ i ═ N). And newly building a set H, putting N frames into the set H, and simultaneously initializing an empty set M for storing the optimal frame.
(1) Sorting the H middle frames box according to the score Si, selecting the highest-score frame box _ M, and moving the box _ M to the set M;
(2) traversing the box in H, respectively calculating an intersection ratio (IOU) with the box _ m, wherein the intersection ratio is expressed as
If the IOU is above a certain threshold, then this box is removed from set H.
(3) And (4) returning to the step (1) for iteration until the set H is empty, and obtaining a final frame selection result by the set M.
Meanwhile, when the deep learning model is trained, a large amount of computing power and time are consumed for reducing the model loss value due to the initialization of random parameters, and the migration learning method is adopted in the embodiment. The convolution layer trained by the transfer learning has the capability of extracting general image features, a pre-training model trained under big data set ImageNet data shares a bottom layer structure weight parameter, and then a top layer network structure is modified, so that the convergence speed can be greatly accelerated in new task training.
The specific method for selecting the pictures of the ore mud block mixture to form the training set and the test set in the model training comprises the following steps:
1) marking the interested region by using the acquired 4600 ore-mud block mixture data set according to the data set format adopted by the SSD official model, and marking the 4-tuple parameter (X) of the targetmin，Ymin，Xmax，Ymax) And respectively representing the coordinates of the upper left corner and the lower right corner of the marking box, and only marking the Mud blocks in the image, namely only two Classes of Mud blocks Mud and Background are available, and the information is stored in an xml file.
2) The training set and the test set are as follows: and 2, dividing according to the proportion, namely, training set counting 3680 and test set counting 920. The training set is used for training network parameters, and the testing set is used for performance evaluation of target prediction.
3) In order to further improve the robustness of the model, the generalization capability of the model is improved by adopting a data enhancement method, and the corresponding transformation function is used for carrying out operations such as turning, cutting, translation and the like on the input image. The data set is input into the network through the iterator, and the iterator performs random image enhancement transformation every time when images are input, so that the data volume is further expanded.
4) The data set is input into the SSD512, and after image data is input into a network, the network is continuously learned, parameters are continuously updated, the regression of positions and the classification accuracy of categories are continuously improved through the feedforward of the network, the combination of a loss function and the SGD gradient descent method, so that an ideal model weight can be finally achieved.
According to the above method, as shown in fig. 2, we easily obtain an ore automatic identification and rough sorting system based on deep learning, which comprises:
the model building module is used for building a deep learning model for automatic ore identification and rough sorting; the deep learning model for ore automatic identification and rough sorting comprises a classification model network and a target detection framework;
the classification model network is a VGG16 network and is used for extracting deep layer network features and shallow layer network features of ores and mud blocks in the training images; the target detection framework is an SSD512 training model and is used for predicting and outputting a mud block recognition result of the ore and mud block mixture according to the deep layer network characteristics and the shallow layer network characteristics of the extracted ore and mud blocks;
the model training module is used for forming a training set and a testing set according to the selected picture of the ore mud block mixture and training a deep learning model for automatic ore recognition and rough sorting;
the model training module selects pictures of the ore mud block mixture to form a training set and a testing set to train the deep learning model of ore automatic identification and rough sorting, and the method specifically comprises the following steps:
1) marking the collected ore and mud block mixture data set according to the data set format adopted by the SSD official model to obtain the target marked 4-tuple parameter (X)min，Ymin，Xmax，Ymax) Respectively representing the coordinates of the upper left corner and the lower right corner of the marking frame, namely only two Classes of Mud blocks Mud and Background are available, and the information is stored in an xml file;
2) the training set and the test set are as follows: 2, dividing in proportion, wherein a training set is used for training network parameters, and a test set is used for performance evaluation of target prediction;
3) inputting the data set into a target detection framework, after image data is input into a network, continuously learning the network, continuously updating parameters, continuously improving the regression of positions and the classification accuracy of categories through the feedforward of the network, combining a loss function and an SGD gradient descent method, and finally obtaining an ideal model weight.
The data set in the step 1) is expanded in the following way: using a transformation function, flipping, cropping, or panning the marked image results in added data.
And 3) carrying out random image enhancement transformation before the image data is input into the target detection frame in the step 3) to obtain increased image data and expand the data volume of the data set.
The recognition module is used for inputting the shot ore-mud block mixture picture on the crawler belt into a trained deep learning model for automatic ore recognition and coarse sorting to obtain a real-time recognition result;
the sorting module is used for sending the ore of the batch into the next procedure if the mud content is less than 3% of the threshold value after the ore is identified according to the identification result of the identification module, otherwise, controlling the high-pressure water gun to carry out fixed-point flushing on the identified mud block and then sending the mud block to the next procedure.
It will be understood that modifications and variations can be made by persons skilled in the art in light of the above teachings and all such modifications and variations are intended to be included within the scope of the invention as defined in the appended claims.
Claims (10)
1. An ore automatic identification and rough sorting system based on deep learning is characterized by comprising:
the model building module is used for building a deep learning model for automatic ore identification and rough sorting; the deep learning model for ore automatic identification and rough sorting comprises a classification model network and a target detection framework;
the classification model network is used for extracting deep layer network features and shallow layer network features of ores and mud blocks in the training image;
the target detection framework is used for predicting and outputting a mud block identification result of the ore and mud block mixture according to the deep layer network characteristics and the shallow layer network characteristics of the extracted ore and mud blocks;
the model training module is used for forming a training set and a testing set according to the selected picture of the ore mud block mixture and training a deep learning model for automatic ore recognition and rough sorting;
the recognition module is used for inputting the shot ore-mud block mixture picture on the crawler belt into a trained deep learning model for automatic ore recognition and coarse sorting to obtain a real-time recognition result;
the sorting module is used for controlling the crawler belt to send the ore of the batch into the next procedure according to the recognition result of the recognition module if the mud content after recognition is smaller than a set threshold value, otherwise, controlling the high-pressure water gun to carry out fixed-point flushing on the recognized mud blocks and then sending the mud blocks to the next procedure.
2. The ore automatic identification and rough sorting system based on deep learning of claim 1, wherein the classification model network in the model construction module is a VGG16 network for extracting deep layer network features and shallow layer network features of ores and mud blocks in training images; and the target detection framework is an SSD512 training model and is used for predicting and outputting the recognition results of the ores and the mud blocks according to the deep layer network characteristics and the shallow layer network characteristics of the extracted ores and mud blocks.
3. The automatic ore identification and rough sorting system based on deep learning of claim 2, wherein the model training module selects pictures of ore mud cake mixture to form a training set and a testing set, and trains the deep learning model of automatic ore identification and rough sorting as follows:
1) marking the collected ore and mud block mixture data set according to the data set format adopted by the SSD official model to obtain the target marked 4-tuple parameter (X)min，Ymin，Xmax，Ymax) Respectively representing the coordinates of the upper left corner and the lower right corner of the marking frame, namely only two Classes of Mud blocks Mud and Background are available, and the information is stored in an xml file;
2) the training set and the test set are as follows: 2, dividing in proportion, wherein a training set is used for training network parameters, and a test set is used for performance evaluation of target prediction;
3) inputting the data set into a target detection framework, after image data is input into a network, continuously learning the network, continuously updating parameters, continuously improving the regression of positions and the classification accuracy of categories through the feedforward of the network, combining a loss function and an SGD gradient descent method, and finally obtaining an ideal model weight.
4. The automatic ore identification and rough sorting system based on deep learning of claim 3, characterized in that the step 1) further comprises a step of expanding the collected ore-mud mixture data set, wherein the data set is expanded by adopting the following method: using a transformation function, flipping, cropping, or panning the marked image results in added data.
5. The automatic ore identification and rough sorting system based on deep learning of claim 3, wherein the image data in step 3) is subjected to random image enhancement transformation before being input into the network to obtain increased image data, thereby expanding the data volume of the data set.
6. An ore automatic identification and rough sorting method based on deep learning is characterized by comprising the following steps:
1) constructing a deep learning model for automatic ore identification and rough sorting; the deep learning model for ore automatic identification and rough sorting comprises a classification model network and a target detection framework;
the classification model network is used for extracting deep layer network features and shallow layer network features of ores and mud blocks in the training image;
the target detection framework is used for predicting and outputting a mud block identification result of the ore and mud block mixture according to the deep layer network characteristics and the shallow layer network characteristics of the extracted ore and mud blocks;
2) selecting pictures of the ore mud block mixture to form a training set and a testing set, and training a deep learning model for automatic ore identification and rough sorting;
3) the method comprises the steps of shooting an ore mud block mixture picture on a crawler belt, inputting the picture into a trained deep learning model for automatic ore identification and coarse sorting, and obtaining a real-time identification result;
4) according to the identification result, if the mud content after identification is less than the set threshold value, the crawler belt is controlled to send the ore of the batch into the next procedure, otherwise, the high-pressure water gun is controlled to carry out fixed-point flushing on the identified mud blocks, and then the mud blocks are sent to the next procedure.
7. The ore automatic identification and rough sorting method based on deep learning of claim 6, wherein the classification model network in the step 1) is a VGG16 network for extracting deep layer network features and shallow layer network features of ores and mud blocks in training images; and the target detection framework is an SSD512 training model and is used for predicting and outputting the recognition results of the ores and the mud blocks according to the deep layer network characteristics and the shallow layer network characteristics of the extracted ores and mud blocks.
8. The automatic ore identification and rough sorting system based on deep learning of claim 7, wherein the model training module selects pictures of ore mud cake mixture to form a training set and a testing set to train the deep learning model of automatic ore identification and rough sorting, which is as follows:
2.1) marking the collected ore and mud block mixture data set according to the data set format adopted by the SSD official model to obtain the target marked 4-tuple parameter (X)min，Ymin，Xmax，Ymax) Respectively representing the coordinates of the upper left corner and the lower right corner of the marking frame, namely only two Classes of Mud blocks Mud and Background are available, and the information is stored in an xml file;
2.2) the training set and the test set are adjusted according to the weight ratio of 8: 2, dividing in proportion, wherein a training set is used for training network parameters, and a test set is used for performance evaluation of target prediction;
and 2.3) inputting the data set into a target detection framework, after image data are input into a network, continuously learning the network, continuously updating parameters, continuously improving the regression of positions and the classification accuracy of categories through the feedforward of the network, combining with a loss function and an SGD gradient descent method, and finally obtaining an ideal model weight.
9. The automatic ore identification and rough sorting system based on deep learning of claim 8, characterized in that the step 2.1) further comprises a step of expanding the collected ore-mud mixture data set, wherein the data set is expanded by adopting the following method: using a transformation function, flipping, cropping, or panning the marked image results in added data.
10. The automatic ore identification and rough sorting system based on deep learning of claim 8, wherein the image data input network in step 2.3) performs random image enhancement transformation to obtain increased image data, expanding the data volume of the data set.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010215037.1A CN111507379A (en) | 2020-03-24 | 2020-03-24 | Ore automatic identification and rough sorting system based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010215037.1A CN111507379A (en) | 2020-03-24 | 2020-03-24 | Ore automatic identification and rough sorting system based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111507379A true CN111507379A (en) | 2020-08-07 |
Family
ID=71864636
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010215037.1A Pending CN111507379A (en) | 2020-03-24 | 2020-03-24 | Ore automatic identification and rough sorting system based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111507379A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112718548A (en) * | 2020-12-10 | 2021-04-30 | 青岛农业大学 | General type agricultural product intelligence sorter of degree of depth study |
| CN113011486A (en) * | 2021-03-12 | 2021-06-22 | 重庆理工大学 | Chicken claw classification and positioning model construction method and system and chicken claw sorting method |
| CN113953067A (en) * | 2021-10-20 | 2022-01-21 | 湖南斯福迈智能科技有限责任公司 | Mineral pre-enrichment method based on solid-state microwave |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20190075707A (en) * | 2017-12-21 | 2019-07-01 | 동의대학교 산학협력단 | Method for sorting products using deep learning |
| CN110135468A (en) * | 2019-04-24 | 2019-08-16 | 中国矿业大学(北京) | A kind of recognition methods of gangue |
| CN110503161A (en) * | 2019-08-29 | 2019-11-26 | 长沙学院 | A kind of ore walk object detection method and system based on Weakly supervised YOLO model |
-
2020
- 2020-03-24 CN CN202010215037.1A patent/CN111507379A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20190075707A (en) * | 2017-12-21 | 2019-07-01 | 동의대학교 산학협력단 | Method for sorting products using deep learning |
| CN110135468A (en) * | 2019-04-24 | 2019-08-16 | 中国矿业大学(北京) | A kind of recognition methods of gangue |
| CN110503161A (en) * | 2019-08-29 | 2019-11-26 | 长沙学院 | A kind of ore walk object detection method and system based on Weakly supervised YOLO model |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112718548A (en) * | 2020-12-10 | 2021-04-30 | 青岛农业大学 | General type agricultural product intelligence sorter of degree of depth study |
| CN113011486A (en) * | 2021-03-12 | 2021-06-22 | 重庆理工大学 | Chicken claw classification and positioning model construction method and system and chicken claw sorting method |
| CN113953067A (en) * | 2021-10-20 | 2022-01-21 | 湖南斯福迈智能科技有限责任公司 | Mineral pre-enrichment method based on solid-state microwave |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111784685B (en) | Power transmission line defect image identification method based on cloud edge cooperative detection | |
| CN111507379A (en) | Ore automatic identification and rough sorting system based on deep learning | |
| CN110263706B (en) | Method for detecting and identifying dynamic target of vehicle-mounted video in haze weather | |
| CN113936256A (en) | Image target detection method, device, equipment and storage medium | |
| CN105791774A (en) | Surveillance video transmission method based on video content analysis | |
| CN107018352A (en) | A kind of method of video image processing, apparatus and system | |
| CN111222478A (en) | Construction site safety protection detection method and system | |
| CN109886159B (en) | Face detection method under non-limited condition | |
| CN112419202B (en) | Automatic wild animal image recognition system based on big data and deep learning | |
| CN110087041B (en) | Video data processing and transmitting method and system based on 5G base station | |
| CN111160389A (en) | Lithology identification method based on fusion of VGG | |
| CN116563293B (en) | Photovoltaic carrier production quality detection method and system based on machine vision | |
| CN113158829A (en) | Deep learning ore size measuring method and early warning system based on EfficientDet network | |
| CN116665092A (en) | Method and system for identifying sewage suspended matters based on IA-YOLOV7 | |
| CN114332749A (en) | Sand production monitoring method of sand production ship based on deep learning | |
| CN101184235B (en) | Method and apparatus for implementing background image extraction from moving image | |
| CN116597411A (en) | Method and system for identifying traffic sign by unmanned vehicle in extreme weather | |
| CN113971666A (en) | Power transmission line machine inspection image self-adaptive identification method based on depth target detection | |
| CN110033045A (en) | A kind of method and apparatus of trained identification image atomization | |
| CN111814726B (en) | Detection method for visual target of detection robot | |
| CN112132207A (en) | Target detection neural network construction method based on multi-branch feature mapping | |
| CN114640785A (en) | Site model updating method and system | |
| CN110674764A (en) | Method, device and system for detecting exposed earthwork of construction site | |
| CN115086663B (en) | Multi-camera real-time video transmission method based on deep reinforcement learning | |
| CN113297909B (en) | Building damage real-time identification method based on unmanned aerial vehicle vision |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |