KR101981031B1 - Platform for scrapping metal based on artificial intelligence - Google Patents

Abstract

Provided is a steel scrapping platform based on artificial intelligence which can quickly and accurately analyze a containing ratio of each component. To this end, according to an embodiment of the present invention, a steel scrapping method comprises the steps of: measuring a weight of a steel product; measuring a volume of the steel product; photographing an image of the steel product; dividing the image into a plurality of fragments by using a grid; analyzing each of the plurality of fragments by using an artificial neural network; obtaining a result that each of the plurality of fragments is classified based on a plurality of classification standards from the artificial neural network; and adjusting the number of fragments when a ratio of the number of fragments having the accuracy in the classified result for the number of all fragments higher than or equal to a preset value is smaller than a preset ratio.

Description

PLATFORM FOR SCRAPPING METAL BASED ON ARTIFICIAL INTELLIGENCE

The present invention relates to an artificial intelligence based iron scrap platform. More specifically, in order to classify iron scrap using artificial intelligence, a steel scrap which can improve the accuracy of analysis by using LiDAR (Light Detection And Ranging) apparatus, a weighing apparatus (weight measuring apparatus) and a video image grid ≪ / RTI >

As information digitization and data storage technology are developed, machine learning technology is introduced and used in various fields. Machine learning is a technique for analyzing a large number of input data, classifying probabilistic objects, or predicting values within a specific range. Machine learning works by a method of empirically analyzing a large number of input data and deriving a result value stochastically rather than deriving a result value by a specific rule. When humans give rules and definitions related to a particular task, the computer processes a large amount of data based on the rules and definitions, and learns how to perform tasks by 'learning' the computer itself.

If the existing machine learning technique is analyzed based on rules and definitions given by human, deep learning, which is a detailed field of machine learning, omits the rules and definitions given by humans in the existing machine learning, Using the neural network consisting of four layers, the computer enables the automation of the tasks that are difficult to express with specific rules or definitions by extracting the characteristics of the objects themselves, analyzing the objects, and giving answers.

Artificial intelligence, that is, image classification, is a representative task performed using machine learning or deep learning. For example, by inputting a large number of labeled images, an artificial neural network can be learned by a map learning method, and an image classification operation can be performed using the learned artificial neural network.

On the other hand, steel scrap means collecting steel products from steel product manufacturing and processing and steel products and re-introducing them into steel products. Recycling of steel resources in countries with limited resources provides high economic value and contributes greatly to environmental conservation.

Currently, iron scrap accounts for more than 95% of the steel resources used in the electric furnace as an important raw material for the steel industry, accounting for more than 50% of the cost of manufacturing steel products. Iron scrap not only can reduce mineral resource consumption, but it also has the advantage that it can obtain high quality steel easily than iron ore from the beginning.

In order to increase the recycling efficiency of steel resources, accurate scrap scrap classification work is required. In the past, a person directly scrapped scrap steel scrap. When the camera photographed a steel product in a lorry in various angles, it was a method in which a person scans an image and performs an iron scrap classification. For this reason, there has been a problem in that a subjective criterion is applied to a conventional scrap iron sorting operation or an accurate classification is not performed according to the state of photographed images.

Therefore, there is a need to perform quick and accurate iron scrap classification work using artificial intelligence. One embodiment of the present invention performs a steel scrap classification by learning LiDAR, an approximation device, and three image images in a complex manner without depending on only a simple image.

Korean Patent Laid-Open No. 10-2015-0123513 (published April 11, 2015)

An object of the present invention is to provide an iron scrap method and apparatus capable of quickly and accurately analyzing the content ratios of respective components constituting an analysis target in an image to be analyzed using artificial intelligence. For example, using artificial intelligence, iron scrap content ratios can be analyzed according to iron scrap classification criteria in steel product images.

Another technical problem to be solved by the present invention is to provide a method and apparatus for analyzing a content ratio of each component constituting an analysis target in an image to be analyzed using artificial intelligence, And to provide an iron scrap method and apparatus. For example, when performing an iron scrap classification in a steel product image using artificial intelligence, a piecewise analysis can be performed to improve the accuracy of the iron scrap classification.

The technical problems of the present invention are not limited to the above-mentioned technical problems, and other technical problems which are not mentioned can be clearly understood by those skilled in the art from the following description.

According to an aspect of the present invention, there is provided a method for scrapping steel, comprising: measuring a weight of a steel product; measuring a volume of the steel product; dividing the image into a plurality of pieces using a grid, analyzing each of the plurality of pieces using an artificial neural network, classifying each of the plurality of pieces into a plurality of classification criteria from the artificial neural network Adjusting the number of pieces when the ratio of the number of pieces with an accuracy of the sorted result to the total number of pieces is equal to or greater than a predetermined value is less than a predetermined ratio, , Verifying the accuracy of the classification result

In one embodiment, each piece of the grid may be rectangular.

In one embodiment, each piece of the grid may differ in transverse aspect ratio.

In one embodiment, adjusting the number of pieces may include increasing the size of each piece of the grid.

In one embodiment, adjusting the number of pieces may include reducing the size of each piece of the grid.

In one embodiment, adjusting the number of pieces may include varying the aspect ratio of each piece of the grid.

In one embodiment, the step of verifying the accuracy may further comprise displaying the grid on the image and displaying the classified results on each piece of the grid.

In one embodiment, the step of displaying the classification result may include displaying the classified results in each piece of the grid by color.

In one embodiment, the image is an image of a steel product loaded on a vehicle, and the plurality of classification criteria may be an iron scrap classification standard.

In one embodiment, the step of obtaining the sorted result may be followed by calculating the iron scrap content ratio according to the iron scrap classification criteria of the steel product using the result of sorting each of the plurality of pieces and the measured volume Step < / RTI >

In one embodiment, the step of verifying the accuracy may further include calculating the price of the steel product using the weight, the content of each scrap, and the unit price of each scrap.

In one embodiment, the step of verifying the accuracy may further include displaying the grid on the image, and displaying the results of the iron scrap classification on each piece of the grid.

In one embodiment, the step of displaying the classified results may include displaying the iron scrap classification results on each piece of the grid by color.

According to another aspect of the present invention, there is provided an apparatus for scraping an iron according to the present invention, comprising: a carpet sensor for measuring the weight of a steel product; a LiDAR sensor for measuring the volume of the steel product; A display for displaying an analysis result of the image, a memory for loading a computer program executed by the processor, one or more processors, and a storage for storing the computer program, An instruction to divide the image into a plurality of fragments using an artificial neural network, an instruction to analyze each of the plurality of fragments using an artificial neural network, an instruction to obtain a result of classifying each of the plurality of fragments into a plurality of classification references from the artificial neural network, , The number of the whole pieces An instruction to adjust the number of fragments and an instruction to verify the accuracy of the classified results using the weight and the volume when the ratio of the number of fragments whose accuracy of the result of the flow is equal to or greater than the predetermined value is less than a predetermined ratio .

In one embodiment, the method may further include instructions to display the image, the grid, and the classified result on the display after an instruction verifying the accuracy.

In one embodiment, the instructions displayed on the display may include instructions to display the classified results in a color-coded manner on the display.

In one embodiment, the image is an image of a steel product loaded on a vehicle with the camera, and the plurality of classification criteria may be an iron scrap classification standard.

In one embodiment, an instruction to obtain the sorted result is followed by calculating the iron scrap content ratio according to the iron scrap classification criteria of the steel product using the result of sorting each of the plurality of pieces and the measured volume Instructions. ≪ / RTI >

In one embodiment, the instruction to verify the accuracy may further include an instruction to calculate the price of the steel product using the weight, the amount of each scrap to be scaled, and the unit price of each scrap.

In one embodiment, the display may further include an instruction to display the classification result according to the image, the grid, and the iron scrap classification criteria on the display after the instruction to verify the accuracy.

In one embodiment, the instruction to be displayed on the display may include an instruction to display a sorting result according to the iron scrap classification criteria on a color-coded display.

According to an embodiment of the present invention, when an image is analyzed using artificial intelligence, a grid is used to divide an image into a plurality of pieces, thereby quickly and accurately analyzing a content rate of each component constituting an analysis target can do.

According to another embodiment of the present invention, when the steel scrap is classified using artificial intelligence, the steel scrap can be analyzed according to the steel scrap classification standard in the steel product image by overlaying the grid on the steel product image to be analyzed.

According to another embodiment of the present invention, when analyzing an image using artificial intelligence, the accuracy of analysis can be improved by dynamically adjusting the number of pieces when the accuracy of the analysis result is less than a predetermined value. For example, in classifying iron scrap using artificial intelligence, if the accuracy of the analysis result is less than a preset value, the accuracy of the iron scrap classification can be improved by dynamically adjusting the number of pieces.

The effects of the present invention are not limited to the effects mentioned above, and other effects not mentioned can be exerted in the technical field of the present invention.

1 is a view illustrating an iron scrap classification image according to an iron scrap manual based on the KS D 2101 standard.
2 is a diagram illustrating an artificial neural network according to an embodiment of the present invention.
3 is a diagram illustrating an image segmentation and an artificial neural network learning process according to an exemplary embodiment of the present invention.
4 is a diagram illustrating an image and a grid to be analyzed according to an embodiment of the present invention.
5 is a view illustrating an iron scrap classification result screen according to an embodiment of the present invention.
6 is a view illustrating an iron scrap classification result screen according to another embodiment of the present invention.
7 is a view illustrating an iron scrap classification process according to an embodiment of the present invention.
FIG. 8 is a diagram illustrating a grid and analysis results using the grid according to an embodiment of the present invention.
9 is a diagram illustrating a modified grid according to an embodiment of the present invention and analysis results using the modified grid.
10 is a view illustrating a specific gravity reference method according to an embodiment of the present invention.
11 is another view illustrating a specific gravity reference method according to an embodiment of the present invention.
12 is a flowchart illustrating an iron scrap method according to an embodiment of the present invention.
13 is a block diagram illustrating an iron scrap apparatus according to an embodiment of the present invention.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS The advantages and features of the present invention and the manner of achieving them will become apparent with reference to the embodiments described in detail below with reference to the accompanying drawings. The present invention may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art. To fully disclose the scope of the invention to those skilled in the art, and the invention is only defined by the scope of the claims. Like reference numerals refer to like elements throughout the specification.

Unless defined otherwise, all terms (including technical and scientific terms) used herein may be used in a sense commonly understood by one of ordinary skill in the art to which this invention belongs. Also, commonly used predefined terms are not ideally or excessively interpreted unless explicitly defined otherwise. The terminology used herein is for the purpose of illustrating embodiments and is not intended to be limiting of the present invention. The singular forms herein include plural forms unless the context clearly dictates otherwise.

The word 'comprises' or 'comprising', as used herein, does not exclude the presence or addition of a component / step / operation other than the mentioned components / steps / operations.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention to be described below relates to a method of analyzing images using artificial intelligence. More specifically, the present invention relates to a method and apparatus for dividing an image into a plurality of pieces using a grid, and feeding back analysis results to adjust the number of pieces, A method and an apparatus capable of accurately analyzing are disclosed.

In the present specification, a description will be given of a process of performing a steel scrap of a steel product using artificial intelligence according to an embodiment of the present invention as an example of a 'one or more component analysis target' A method and apparatus for scraping iron using artificial intelligence according to an embodiment of the present invention will be described.

Prior to the description of the present specification, some terms used in this specification will be clearly described.

In the technical field to which the present invention pertains, 'raw steel scrap' refers to iron scrap generated in a steel making process using iron ore as a raw material, and unfouled steel scrap generated in the primary processing of iron. That is, steel material that is plated, painted, or not excessively oxidized, and whose origin or place of origin is clear.

In the technical field to which the present invention belongs, "scrap" refers to steel scrap which has already been treated as being useless, and which is processed to be suitable for reuse. Scrap is generated from waste vehicles, equipment, railways, machinery, ships, building materials, destruction works, and waste products of other consumer / industrial products. Since the sources and areas are scattered widely, , Uneconomical iron scrap is often avoided to collect and become waste.

In the technical field to which the present invention belongs, 'shelf scrap' means a scrap of scrap formed in the machining process.

In the technical field to which the present invention pertains, "processed scrap" means iron scrap generated in the process of manufacturing industrial or consumer products using steel in a machine factory, a steel processing factory, a shipbuilding / automobile factory, Scrap is produced in various forms depending on the equipment used in the steel consuming industry and the manufacturing process.

In the technical field to which the present invention belongs, "heavy scrap" refers to a steel scrap having a relatively large specific gravity to volume, such as rods, sections, mechanical structures, and pipes.

In the technical field to which the present invention belongs, 'lightweight scrap' is a relatively lightweight steel scrap, which includes lightweight structures, life scrap metal and the like.

In the technical field to which the present invention belongs, 'neural network' means a graph structure composed of a plurality of layers made up of a biological neural network, especially a human visual / auditory cortex, and a plurality of nodes constituting each layer . The artificial neural network includes one input layer, one or more hidden layers, and one output layer.

In the technical field of the present invention, 'input layer' means a layer receiving data to be analyzed / learned in a layer structure of an artificial neural network.

In the technical field of the present invention, an 'output layer' means a layer in which a result value is output in a layer structure of an artificial neural network.

The 'hidden layer' in the technical field of the present invention means all layers excluding the input layer and the output layer in the layer structure of the artificial neural network. Artificial neural networks consist of successive layers of neurons, with neurons in each layer connected to neurons in the next layer. If the input layer and the output layer are directly connected, each input contributes independently to the output regardless of the other input, and it is difficult to obtain an accurate result. In the real world, one or more input data are interdependently combined with each other to affect the output in a complex structure, so that this hidden layer can be added so that hidden neurons can capture subtle interactions between inputs that affect the final output. In other words, it can be seen that the hidden layer processes high-level features and attributes in the data.

In the technical field to which the present invention belongs, "neuron" is a concept in which a neuron in a biological neural network is associated with an artificial neural network. The term 'neuron' is also referred to as a 'node'. In the present specification, a 'neuron' is referred to as a 'node' for convenience.

In the technical field to which the present invention belongs, "weight" is a concept in which the degree of enhancement of synapse connection through repetitive signal transmission in a biological neural network is associated with an artificial neural network. In biological neural networks, when signals are transmitted from neuron 1 to neuron 2 frequently, the pathway from neuron 1 to neuron 2, the synaptic connection, is strengthened for signaling efficiency. This is a kind of learning or memory process, and artificial neural networks describe it as 'weight'.

BACKGROUND OF THE INVENTION Field of the Invention The present invention relates to a back propagation method and a back propagation method for detecting a back propagation of an error, Which means adjusting the weight of the artificial neural network.

Convolutional Neural Networks (CNN) is a deep learning model that combines artificial neural networks with filter technology, and has been optimized so that artificial neural networks can acquire better characteristics of input data . Convolutional artificial neural networks perform well in the field of computer vision.

In the present specification, the term " grid " means an aggregate formed by intersecting a plurality of lines. For example, in the case of a rectangular grid, a plurality of vertical lines define a vertical column, and a plurality of horizontal lines define a horizontal row. The plurality of vertical lines and the plurality of horizontal lines form a plurality of spaces isolated from each other. The grid may be a rectangular grid or a polygonal grid such as a hexagon. However, for the sake of convenience of description, a rectangular grid will be described in this specification.

As used herein, the term " piece " means an area or an image within each area divided by a plurality of lines constituting the grid. The grid consists of a plurality of pieces. The piece may have a rectangular shape, or may be a polygonal shape such as a hexagon.

In the present specification, 'object to be analyzed' means an object to be analyzed using an artificial neural network according to an embodiment of the present invention. The artificial neural network according to an embodiment of the present invention can analyze the ratio of each constituent element constituting the analysis object. For example, the analysis target may be a steel product to be subjected to iron scrap for recycling, and the artificial neural network according to an embodiment of the present invention may analyze the content of iron scrap according to the steel scrap classification standard have.

In the present specification, an 'analysis target image', a 'target image', or an 'image' refers to an image of an analysis target. For example, the image to be analyzed may be an image of a steel product photographed by a camera.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.

Iron scraps are classified by source, ingredient, form, purchase type and so on. Among them, iron scrap classification standard based on KS D 2101 standard, which is published in the iron scrap manual distributed by the Iron Scrap Committee of Korea Iron & Steel Association, is as follows.

According to KS D 2101 standards, scrap scraps are classified as shown in Table 1.

Classification Representative product type Fresh Iron A (KA1) - hot rolled plate
- Hot blast furnace and electric furnace hot rolled, heavy plate and punching,
- Plated and unpainted pipes before use
- Bonding Good wire, coil scrap of coil
- Cut slab scrap, side scrap (side trimming and side turning scrap) Fresh Iron B (KA2) - cold rolled plate, sheet cutting and punching, silicon steel plate
- Ferrous scrap of coils with good binding
- Side scrap (Side trimming and side scoring) Blood Pressure (KA3) Compressed raw material base material is 600mm or less in width, height and height (tolerance of each width, height, height should be within 10%) Beef L (KA4) For raw meat A and raw meat B, but with a length exceeding 1,000 mm and a width of 500 mm,

According to KS D 2101 standard, scrap is classified as shown in Table 2.

Motor block
(KBM) Length ≤1,000㎜, Width ≤500㎜, Weight ≤600㎏ Only for motor block generated in heavy equipment of automobile. Weight A (KB1) - Before using straight steel, forging scrap, large vehicle housing, large vehicle frame, waste casting,
- cast iron, molds, waste cast steel,
- Used pipe of Ø300mm or more
- Cutting high pressure gas cylinder (LPG cylinder excluded), plate material welding structure
- Waste ships, heavy equipment disassembly, large vehicles and train wheels
- Bolts and nuts (punching scrap), springs and chains
- Steel bars (round steel, flat steel, angles, etc.)
- Shapes (H, I, a, c, T, except C, steel)
- bearings, billets, sheet piles, squeeze steel pipes
- Iron scrap, such as rail related (rail, wheel, axle, anchor, shoulder, rail under plate by forging or rolling) Weight B (KB2) - Reinforcement after use, cramp (unused)
- Small vehicle housing, compact car wheel
- Waste ships (other than weight A) Plate-type scrap generated during the dismantling process
- incised LPG barrel, incision well cast iron pipe
- Used Ø50 ~ 300mm cut pipe and support pipe, pipe connection bracket
- Iron scrap of various machinery and parts Weight AL (KB31) It is a weight A class, but it has a length of 1,000 mm, a width of 500 mm and a weight of more than 600 kg. Weight BL (KB32) Weight B, which corresponds to the classification, but has a length of 1,000 mm, a width of 500 mm, a weight of 600 kg or more and a weight B Lightweight A (KB4) - Used pipes less than Ø50 mm,
- C-shaped steel, angle, color steel plate, wire rope
- Small vehicle frame, steel banding, transformer case
- Steel Euro Form, Steel Form,
- Iron scrap of construction waste materials (window, door and shutter, steel frame, window, etc.) Lightweight B (KB5) - retirement sandwich panels, steel toys, umbrella stands
- muffler, tin plate, tin, lightweight structure, wire, wire mesh, cartridges, incised drums, bicycle
- Household goods and office tools
- Home appliances case (microwave, cold damp, washing machine, etc.)
- Office appliances (cabinets, steel pile boxes, desk chairs, etc.)
- Heating equipment (stove, dismantling furnace, etc.)
- Iron scrap, such as iron scrap, which is generated in daily life (except for tableware) Lightweight L (KB6) - Lightweight A and lightweight B, but with a length of 1,000 mm, a width of 500 mm and a weight of more than 600 kg,

Shelf scrap according to KS D 2101 is classified as shown in Table 3.

Classification Representative product type Shelf A (KC1) Scrap and clip-like scrap generated during machining, thickness ≥1 mm Shelf B (KC2) Scrap generated during main product lathe processing Shelf C (KC3) Scrap and clip-like scrap generated during machining, thickness <1mm

According to KS D 2101, processed scraps are classified as shown in Table 4.

Classification Representative product type Shredder A (KS1) - Iron scrap processed into shredder using raw iron as base metal
- width ≤200mm, length and height ≤100mm Shredder B (KS2) - Charcoal, panda, clipping, etc., of scrapped and refined cars that do not contain non-ferrous metals
- width ≤200mm, length and height ≤100mm Shredder C (KS3) Light scrap shredded scrap
Width ≤200mm, length and height ≤100mm Guilotin A
(KG1) The material characteristic is light A or more, and scrap with cutting with Gilotin shear
Width ≤1,200mm, Length ≤700mm Guilotin B
(KG2) Light material BDP corresponds to the material characteristics, scrap with cutting with Gilotin shear
Width ≤1,200mm, Length ≤700mm Compression A
(KP1) Compressed with base metal of light A or more
Horizontal, vertical, and height of 600mm or less, respectively (The tolerance of each width, height, height should be within 10%)
Weight ≤600kg Compression B (KP2) Compressed lightweight B with base metal
Horizontal, vertical, and height of 600mm or less, respectively (The tolerance of each width, height, height should be within 10%)
Weight ≤600kg Compression C (KP3) Compression of shelf scrap B with base metal
Horizontal, vertical, and height of 600mm or less, respectively (The tolerance of each width, height, height should be within 10%)
Weight ≤600kg Compression D (KP4) Compression of shelf scrap C with base metal
Width, height, height not more than 600mm each (tolerance of each width, height, height should be within 10%)
Weight ≤600kg Compression E (KP5) Compressed can with base metal
Tolerance of each width, height, height should be within 10%)
Weight ≤600kg

The iron scrap manual distributed by the Iron Scrap Committee of the Korea Iron & Steel Association provides representative iron scrap images according to each classification.

1 is a view illustrating an iron scrap classification image according to an iron scrap manual based on the KS D 2101 standard.

FIG. 1 (a) is an illustration of a plate scrap image having a thickness of 3.0 mm and a length of? 1,000 mm corresponding to 'raw iron A (KA1)', mm &lt; / RTI &gt; thickness &lt; 3.0 mm, length &lt;

Fig. 1 (c) is an illustration of a raw-frozen compressed scrap image of horizontal, vertical, and height ≤ 600 mm corresponding to the 'barometric pressure (KA3)'. Fig. 1 (d) shows a slab scrap having a thickness ≥3.0 mm and a length ≤1,000 mm Image.

As shown in FIG. 1, the iron scrap has various characteristics such as line, surface, shape, color, and size according to its kind. However, it has an irregular shape which is difficult to classify iron scrap by analyzing the characteristics of iron scrap image by rule based method. However, when a person directly carries out the iron scrap classification, there is a problem that a subjective judgment criterion is applied or an accurate classification is not performed due to the state of the image and the worker condition. Accordingly, an embodiment of the present invention discloses a method and apparatus for performing steel scrap classification using a deep running technique.

2 is a diagram illustrating an artificial neural network according to an embodiment of the present invention.

In image classification using machine learning, the performance of image classification depends on how much good features are extracted from the input image. In the conventional machine learning method, learning is performed in the order of "input → feature extraction → result" through an intermediate step of feature extraction. For example, in order to recognize an object in a photograph, first a characteristic line or a characteristic color distribution is first extracted from a pixel value, and then a determination is made based on the extracted characteristic line or a characteristic color distribution.

Deep learning is a method of machine learning with a deep structure, which performs learning without extracting features from the input image. Deep learning can perform an operation of extracting features using an artificial neural network including one input layer, one output layer, and one or more hidden layers as part of the overall learning algorithm, and the deep structure of the artificial neural network It allows you to process your work properly by reflecting variables.

The artificial neural network that constitutes the deep learning consists of a node and a synapse that connects the node. The synapses have weights, which are estimated based on function values that project the values learned by the data. When one node group is activated and weights from the data are reflected to other nodes, the last node activated becomes the classification result.

In the conventional machine learning method, a preprocessing process (PCA) is performed to extract features to be matched to each number, and then, based on the processed features, Respectively. However, deep learning also learns how to extract optimal features from raw data, which is called representation learning or feature learning. When a person distinguishes the number '2', he considers '2' to be '2'. In other words, if we continue to look at '2' in various shapes, we will recognize the feature of '2', so that we can distinguish '2'. Deep learning analyzes the object in a similar way as a person distinguishing objects, in which the in-depth structure of artificial neural networks and weight adjustment techniques are used. In the human brain, when the same stimulus is repeatedly transmitted or a reward prediction error occurs, the synaptic weights are changed in such a way that more neurotransmitters are secreted to the same stimulus or more receptors are formed in the dendrites . An artificial neural network similarly adjusts the weights of synapses that connect the nodes in a backpropagation manner, thereby providing more accurate output values.

CNN (Convolutional Neural Networks) 'is one of the deep running techniques that is specialized in image analysis and combines image filtering technology with artificial neural networks. Convolutional artificial neural networks perform well in the field of computer vision.

The present invention relates to an iron scrap method and apparatus capable of quickly and accurately analyzing the content ratios of respective components constituting an analysis target in an image to be analyzed using artificial intelligence, and it is an object of the present invention to use CNN desirable. However, the depth learning technique used in the present invention is not limited to the CNN technique, and any technique may be used as long as it is a technique using an artificial neural network having a deep structure.

3 is a diagram illustrating an image segmentation and an artificial neural network learning process according to an exemplary embodiment of the present invention.

A sufficient learning image is required for building an artificial neural network according to an embodiment of the present invention. For example, for steel scrap of steel products, it is necessary to learn each artificial neural network by inputting each image region classified according to each steel scrap classification in the image of steel product into artificial neural network.

However, as shown in Fig. 3, not all portions of the images of steel products are suitable for artificial neural network learning. The image of steel products contains foreign substances such as dust, which are difficult to be classified as specific scrap iron, and shadows that are difficult to distinguish from the outline.

Therefore, it is desirable to separate each image area that can be distinguished according to each iron scrap classification from an image of a steel product, input labels into the artificial neural network by attaching a label of each iron scrap classification result to each separated image. At this time, the image used for learning of the artificial neural network is preferably a lossless image format.

Google 's Deep Learning Engine requires about 5,000 learning images per figure to identify the numbers. Accordingly, in artificial intelligence learning for image analysis according to an embodiment of the present invention, it is preferable to perform map learning by inputting at least 3,000 pictures, preferably about 5,000 pictures, for each iron scrap classification.

The artificial neural network learning process includes the steps of analyzing each image region classified according to each iron scrap classification in an image of a steel product loaded on a vehicle using an artificial neural network, Comparing the classification value of the node with an actual classification value attached to the label of the input image to obtain an error, and modifying the weight of the artificial neural network by back propagating the delta.

When the artificial neural network for the iron scrap classification is sufficiently learned by using the deep running technique, for example, the CNN technique, the iron scrap classification work using the artificial intelligence can be performed.

4 is a diagram illustrating an image and a grid to be analyzed according to an embodiment of the present invention.

As shown in FIG. 4, a grid consisting of a plurality of vertical lines and a plurality of horizontal lines is overlaid on the image to be analyzed, in order to analyze a content ratio of each component constituting the analysis target in the image to be analyzed. A plurality of vertical lines define a vertical column, a plurality of horizontal lines define a horizontal row, and a plurality of vertical lines and a plurality of horizontal lines form a plurality of isolated spaces, that is, a plurality of pieces.

Overlaying the grid on the image to be analyzed in order to analyze the content ratio of each component constituting the analysis target in the image to be analyzed does not mean displaying the grid visually on the image, To obtain a plurality of pieces by dividing them into units.

When the grid is overlaid on the image to be analyzed, the image of the part corresponding to each piece area of the grid becomes a piece. That is, the analysis target image can be divided into a plurality of pieces using the grid.

FIG. 4 illustrates a case where iron scrap of a steel product is performed using an artificial neural network according to an embodiment of the present invention when the image to be analyzed is an image of a steel product. The grid illustrated in FIG. 4 is composed of 10 pieces of 10 * 10 = 100 pieces in 10 rows and 10 rows. Therefore, the number of fragments that can be obtained from the image to be analyzed using the grid illustrated in Fig. 4 becomes 100.

As shown in FIG. 4, in the iron scrapping method according to the embodiment of the present invention, each of a plurality of pieces corresponding to each sculptural area is input to an artificial neural network without inputting the entirety of the analysis target image into the artificial neural network. For example, after an image of a steel product is divided into 100 pieces by using a grid, each piece is inputted to an artificial neural network to perform iron scrap classification for each piece.

When a piece is input to the input layer of the artificial neural network, the probability that the piece corresponds to each iron scrap classification in the output layer of the artificial neural network is outputted as a certain range of numbers through the weight calculation process by the hidden layer of the artificial neural network. The value output to each node of the output layer of the artificial neural network may be a value between 0 and 1. Alternatively, the range of the output value can be set to any range of numbers according to the user setting. For example, the value output to each node of the output layer of the artificial neural network can be set to a value between 0 and 100 (percent value).

For example, when the output value of each node of the output layer of the artificial neural network is set to a value between 0 and 100 percent, a piece of the steel product image is input to the artificial neural network. As a result, The value of the node representing light weight A is 10%, the value of node representing light weight A is 10%, the value of node representing light weight B is 65%, and the value of dust Is 10%, it can be judged that the result of the scrap classification of the piece is 'light weight B' having the largest value. In this way, iron scrap classification is performed for each piece of steel product image divided by the grid.

On the other hand, if the value of the node having the largest value output from the output layer of the artificial neural network as a result of inputting the fragment into the artificial neural network is a predetermined value, for example, 50% or less, It can be judged. That is, when the value of the node having the largest value in the output layer of the artificial neural network is less than a preset value, it can be judged that the sculpture can not be classified correctly.

If the ratio of fragments which can not classify the accurate scrap scrap to the ratio of the image to be analyzed is more than a predetermined ratio, it can be judged that the accurate scrap scrap classification is impossible with the current grid. In this case, a piecewise analysis can be performed to improve the accuracy. This will be described later in detail with reference to FIGS. 6 to 9.

As described above, the result of the scrap classification of each piece is determined by the node having the highest value in the output layer of the artificial neural network. However, when the value of the node having the largest value is equal to or less than a predetermined value, it can be determined that iron scrap classification for the piece is impossible.

Iron scrap classification results can be displayed as letters in each piece. For example, by displaying a letter &quot; A &quot; in a piece including a piece classified as &quot; weight A &quot;, and displaying a letter &quot; B &quot; in a piece containing a piece classified as a weight B, The classification result can be displayed.

Alternatively, the results of the scrap classification can be displayed in the color of each piece. For example, fragments classified as 'heavy A' are displayed in green, fragments classified as 'weight B' are displayed in yellow, fragments classified as 'light B' are displayed in orange, and fragments classified as 'light L' The pieces can be displayed in gray, and the pieces classified as dust can be displayed in red.

On the other hand, the iron scrap price of steel products can be calculated using the results of scrap iron scrap classification. For example, for the total number of pieces, the ratio of the number of pieces classified as weight B to 43.4%, the ratio of the number of pieces classified as weight A to 28.3%, the ratio of the number of pieces classified as light B Is 15.3%, the ratio of the number of pieces classified as lightweight L is 7.3%, and the ratio of the number of pieces classified as foreign matter is 5.7%.

Assuming that the weight of steel products is 10t and the weight per area of each steel scrap is the same, the weight of iron scrap classified as weight B is 4.4t, the weight of scrap classified as weight A is 2.8t , The weight of iron scrap classified as lightweight B is 1.5t, the weight of iron scrap classified as lightweight L is 0.7t, and the weight classified as dust can be estimated as 0.6t. You can calculate the steel scrap price of an entire steel product by multiplying the unit price per kilogram of each steel scrap class for each weight.

5 is a view illustrating an iron scrap classification result screen according to an embodiment of the present invention.

FIG. 5 is a result of performing iron scrap classification using the results obtained by inputting each piece of the analysis object image of the steel product of FIG. 4 into the artificial neural network.

As shown in FIG. 5, as a result of the artificial neural network analysis, the analysis target image in FIG. 4 was analyzed to be composed of 97% of iron scrap A and 3% of iron B scrap.

As described above, the iron scrap classification result can be displayed in color so that the user can easily confirm it. For example, a weight A of iron scrap may be displayed in purple, and a weight B of scrap may be displayed in orange.

Alternatively, the results of scrap iron scrapping can be displayed in a pattern for easy identification by the user. For example, a weight A of iron scrap can be displayed in a dot pattern, and a weight B of iron scrap can be displayed in a hatched pattern.

6 is a view illustrating an iron scrap classification result screen according to another embodiment of the present invention.

FIG. 6 is a result of performing scrap classification using the results obtained by inputting each piece of the analysis object image of another steel product into the artificial neural network.

As shown in FIG. 6, analysis of the artificial neural network results showed that the image of the analysis of the other steel products was composed of 93% of the iron scrap of Guilin A and 7% of the iron scrap of Guilotin B.

As described above, the iron scrap classification result can be displayed in color so that the user can easily confirm it. For example, iron scrap of guillotine A may be displayed in green and iron scrap of guillotine B may be displayed in yellow.

Alternatively, the results of scrap iron scrapping can be displayed in a pattern for easy identification by the user. For example, iron scrap of guillotine A can be displayed in dot pattern, and iron scrap of guillotine B can be displayed in X pattern.

Once the analysis of the types and ratios of iron scrap from the image to be analyzed is completed, the prices of the steel products can be calculated using the types and ratios of the analyzed scrap iron.

Hereinafter, a concrete method of calculating the prices of steel products using the types and ratios of the analyzed scrap iron will be described.

7 is a view illustrating an iron scrap classification process according to an embodiment of the present invention.

As shown in FIG. 7, the steel material loaded on the vehicle is photographed with the LiDAR sensor, a camera, or the like at the top of the vehicle in a stacked state. At this time, the photographed image is referred to as a primary image or a top image.

FIG. 7 is a graph showing the results of the classification of the iron scrap in the primary image using the artificial neural network according to an embodiment of the present invention and the volume measured by the LiDAR sensor. As a result, the weight A ratio is 50% , The proportion of the weight B is 20%, the proportion of the lightweight B is 16%, the proportion of the lightweight L is 8%, and the proportion classified as the impurity is 6%.

Once the iron scrap classification of the primary image is completed, the upper steel is removed using a magnetic crane. Take out the steel at the top, and then take the steel that has been loaded at the break with a LiDAR sensor or camera. At this time, the photographed image is referred to as a secondary image or an interrupted image.

FIG. 7 is a graph showing the results of the classification of iron scrap in the secondary image using the artificial neural network according to an embodiment of the present invention and the volume measured by the LiDAR sensor. As a result, the weight A ratio is 40% , The proportion of the weight B is 30%, the ratio of the lightweight B is 20%, the proportion of the lightweight L is 6%, and the proportion of the impurities is 4%.

Once the iron scrap classification of the secondary image is complete, a magnetic crane is used to remove the steel from the break. After removing the steel material from the suspension, the steel material placed on the lower end is taken with a LiDAR sensor or a camera. The photographed image is referred to as a tertiary image or a lower image.

FIG. 7 is a graph showing the results of analysis of the steel scrap of the tertiary image using the artificial neural network according to an embodiment of the present invention and the volume measured by the LiDAR sensor. As a result, the weight B ratio is 40% , The weight A ratio is 35%, the light weight ratio B is 10%, the light weight L ratio is 8%, and the proportion classified as impurities is 7%.

When the image analysis is completed for 3 times, the classification values of the primary (upper) image, the secondary (suspended) image, and the tertiary (lower) image are averaged to derive the analytical average value three times. The average ratio of the weight B to the total steel product is (50 + 40 + 35) / 3 = 41.7% and the average ratio of the weight B is (20 + 30 + 40) / 3 The average ratio of lightweight L is (8 + 6 + 8) / 3 = 7.3%, the average ratio of lightweight B is (16 + 20 + 10) / 3 = 15.3% The ratio is (6 + 4 + 7) / 3 = 5.7%.

Assuming that the weight of the steel product photographed by the image is 10t and the weight per area of each scrap is the same, the weight of the scrap classified as weight A is 4.2t and the weight of the scrap classified as weight B , The weight of iron scrap classified as lightweight B is 1.5t, the weight of iron scrap classified as lightweight L is 0.7t, and the weight classified as impurity is estimated to be 0.6t. You can calculate the steel scrap price of an entire steel product by multiplying the unit price per kilogram of each steel scrap class for each weight.

Although FIG. 7 exemplifies the case of analyzing the primary (upper) image, the secondary (suspended) image, and the tertiary (lower) image three times, the number of image analysis times may be changed. In other words, the image can be analyzed more than three times by reducing the amount of steel that is removed by adjusting the magnetic crane, which can improve the accuracy of the analysis. For example, if four times of steel material are taken out using a magnetic crane, five (5) times of images are taken in the order of primary (upper) image, secondary (middle), tertiary Bottom) image, and in this case, the accuracy of the classification can be improved compared to analyzing the image three times.

FIG. 8 is a diagram illustrating a grid and analysis results using the grid according to an embodiment of the present invention.

The grid illustrated in Fig. 8 is the same as the grid illustrated in Fig. The grid illustrated in Fig. 8 is composed of 10 pieces of 10 pieces of 10 pieces, 10 pieces of pieces of 10 pieces of 10 pieces of pieces.

The numbers shown in the grid of FIG. 8 are the results of performing artificial neural network analysis using the grid of FIG. When an image of a piece is input to the input layer of the artificial neural network, the probability of the corresponding piece in the output layer of the artificial neural network corresponding to each iron scrap classification is outputted as a certain range of numbers through an operation process by a hidden layer of the artificial neural network . The value output to each node of the output layer of the artificial neural network may be a value between 0 and 1, or a range of output values may be defined by a range of numbers according to user setting. For example, it is also possible to set the value output to each node of the output layer of the artificial neural network to a value (percentage value) between 0 and 100. In the following description, however, a description will be given on the premise that a value between 0 and 1 is output for convenience of explanation.

The analysis target image is divided into 100 pieces using a grid of 10 * 10, and the image of each piece is input to the artificial neural network and the result is shown in FIG. The color of each piece represents the iron scrap classification with the highest output value in that piece, and the numbers listed in each piece represent the highest output value of that piece.

As a result of artificial neural network analysis, the number of pieces classified as iron scrap A is 34, the number of pieces classified as iron scrap B is 41, the number of pieces classified as iron scrap C is 15, The number of unclassified fragments that can not be classified as iron scraps is 10.

8, when artificial neural network analysis is performed using a grid of 10 * 10, the number of pieces having an output value of 0.9 or more is 86. That is, when artificial neural network analysis is performed using a grid of 10 * 10, it can be seen that 86/100 = 86% of the pieces have an accuracy of 0.9 or more.

The allowable level at which the user judges a correct iron scrap classification can be arbitrarily designated according to the user. For example, a tolerance level judged by a user as an accurate iron scrap classification may be designated as a case where the proportion of pieces having an accuracy of 0.9 or more is 90% or more.

Assuming that the predetermined allowable level is 90% of the fragments having an accuracy of 0.9 or more, the result of performing artificial neural network analysis using the grid of 10 * 10 in FIG. 8 indicates that the ratio of fragments having an accuracy of 0.9 or more Is 86%, it can be seen that the result is below the permissible level.

When the result of the artificial neural network analysis is below the allowable level, the iron scrap method according to an embodiment of the present invention can perform the artificial neural network analysis again by adjusting the grid.

For example, grid adjustment may be to loosely configure the grid to increase the size of each piece. For example, the size of each piece may be reduced to four times the size of the grid by 5 * 5, and then input to the artificial neural network to perform artificial neural network analysis again.

As another example, grid adjustment may be to compact the grid to reduce the size of each piece. For example, it is possible to reduce the size of each piece by a factor of 20 * 20, reduce the size of each piece by a factor of four, input it into the artificial neural network, and perform artificial neural network analysis again.

Hereinafter, the case where the grids are denser, the size of each piece is reduced, and then the iron scrap classification is performed by inputting to the artificial neural network will be described.

9 is a diagram illustrating a modified grid according to an embodiment of the present invention and analysis results using the modified grid.

The modified grid illustrated in FIG. 9 is composed of 20 pieces of 20 × 20 = 20 pieces of 20 × 20 pieces.

The numbers in the grid of FIG. 9 are the result of applying artificial intelligence. When the image of the piece is input to the input layer of the artificial neural network, the probability that the piece corresponds to each iron scrap classification in the output layer of the artificial neural network is outputted as a certain range number through the operation process by the hidden layer of the artificial neural network , The value output to each node of the output layer of the artificial neural network may be a value between 0 and 1.

Figure 9 shows the result of analyzing the image to be analyzed by dividing it into 400 pieces using a grid of 20 * 20 and inputting the image of each piece into the artificial neural network. The color of each piece represents the iron scrap classification with the highest output value in that piece, and the numbers listed in each piece represent the highest output value of that piece.

As a result of artificial neural network analysis using a grid of 20 * 20, the number of pieces classified as iron scrap A was 136, the number of pieces classified as iron scrap B was 164, The number of unclassified fragments classified as dust and scrap is 40 pieces.

9, when an artificial neural network analysis is performed using a 20 * 20 grid, the number of pieces having an output value of 0.9 or more is 362. That is, when artificial neural network analysis is performed using a grid of 20 * 20, it can be seen that all the pieces of 362/400 = 90.5% have an accuracy of 0.9 or more.

Assuming that the predetermined allowable level is 90% of the fragments having an accuracy of 0.9 or more, the result of performing artificial neural network analysis using the grid of 20 * 20 in FIG. 9 is that the ratio of fragments having an accuracy of 0.9 or more And 90.5%, respectively. Therefore, the analysis can be ended without any further grid adjustment and the iron scrap classification result can be confirmed.

Meanwhile, according to one embodiment of the present invention, the accuracy of the artificial intelligence based iron scrap analysis can be improved by using the specific gravity of the iron scrap.

Iron scrap is a kind of garbage that is generated in the industrial field and is often not uniform in shape. Therefore, it is difficult to classify iron scrap accurately by image analysis alone.

In order to overcome this problem, the accuracy of the analytical result is determined by using the weight of each steel scrap. When it is judged that the accuracy is lowered by using the specific gravity, the artificial intelligence analysis is performed again, . That is, if the artificial intelligence described above is used to perform iron scrap using only visual information, additional accuracy is determined using specific gravity information according to the type of iron scrap, and artificial intelligence analysis is performed again when the accuracy is low, Accuracy can be improved.

LiDAR sensors can be used to measure the volume of steel products loaded on trucks. Using a LiDAR sensor at the top of the truck, the laser is emitted to measure the distance, so you can see the height of the steel product loaded on the truck and use it to calculate the volume of the steel product.

We can also measure the weight of a steel product by measuring the weight of a truck carrying a steel product and the weight of a truck that is not loaded with a gauge device.

10 is a view illustrating a specific gravity reference method according to an embodiment of the present invention.

Fig. 10 (a) illustrates the height of each section measured using LiDAR, and Fig. 10 (b) illustrates the volume of each piece derived using the height of each section.

Suppose that the steel product consists of only one type of scrap, and that the type of scrap is judged to be lightweight A as a result of the artificial intelligence image analysis. Since steel scrap can have various shapes and voids depending on the discharged industrial field and is not necessarily made up of the same elements, the exact weight of steel scrap can not be calculated and the specific gravity should be determined statistically. Statistically, the weight of lightweight A is 1.5.

In Fig. 10 (b), the total volume of steel products is 1.6 + 2.2 + 2.1 + 1.7 + 1.1 + 1.5 + 2.5 + 1.7 + 2.8 = 17.2 m³. . Since the specific gravity is a relative value to the density of water and the density of water is 1000 kg / m³, the weight of steel products should be 17.2 * 1.5 * 1000 = 25800 kg = 25.8 tons if the iron scrap classification is correct to be lightweight A. If the error range is 1 ton and the weight of steel products is 26 tons, it can be judged as an accurate scrap within the error range. However, if the weight of the steel product is 17 tons, it can be assumed that the steel product is steel scrap with a smaller weight than the lightweight A. Therefore, in this case, the artificial intelligence image analysis is repeatedly performed until the iron scrap having a smaller specific gravity than the lightweight A is derived, so that the accuracy of the iron scrap can be improved by considering the three-dimensional weight information together with the visual analysis.

11 is another view illustrating a specific gravity reference method according to an embodiment of the present invention.

Fig. 11 (a) illustrates the height of each section measured using LiDAR, and Fig. 11 (b) illustrates the volume of each piece derived using the height of each section. In Fig. 11 (b), the pattern of each piece indicates the kind of iron scrap.

Assume that steel products consist of only three kinds of scrap iron, and that the types of scrap iron are lightweight A, shelf A, and compression A as a result of artificial intelligence image analysis. Since steel scrap can have various shapes and voids depending on the discharged industrial field and is not necessarily made up of the same elements, the exact weight of steel scrap can not be calculated and the specific gravity should be determined statistically. Statistically, let's assume that the specific gravity of lightweight A is 1.5, the specific gravity of shelf A is 1.2, and the specific gravity of compression A is 2. The light weight A is a portion indicated by a checkered pattern, the shelf A is a portion indicated by a dot pattern, and the compression A is a portion indicated by an oblique line.

In Figure 11 (b), the volume of lightweight A is 1.6 + 2.2 + 2.1 + 1.7 = 7.6 m³, the weight of shelf A is 1.1 + 1.5 = 2.6 m³ and the weight of compression A is 2.5 + 1.7 + 2.8 = 4.2 m³. When we calculate the specific gravity of each steel scrap, the weight by light weight A is 7.6 * 1.5 * 1000 = 11400 kg, the weight by shelf A is 2.6 * 1.2 * 1000 = 3120 kg and the weight by compression A is 4.2 * 2 * 1000 = 8400 kg. Therefore, if the iron scrap classification is correct, the weight of the entire steel product should be 22.82 tons. If the error range is 1 ton and the weight of the steel product is 23 tons, it can be judged as an accurate scrap within the error range. However, if the weight of steel products is 27 tons, it can be inferred that iron scrap of steel products is misdetected. Therefore, in this case, when the weight is considered together, artificial intelligence image analysis is repeated until accurate steel scrap is derived, so that accuracy of scrap can be improved by considering stereoscopic weight information in visual analysis.

12 is a flowchart illustrating an iron scrap method according to an embodiment of the present invention.

As shown in FIG. 12, the iron scrap method according to an embodiment of the present invention includes steps of photographing an image (S1210), dividing an image into a plurality of pieces using a grid (S1220) A step S1230 of analyzing the plurality of pieces of the plurality of pieces from the artificial neural network using the artificial neural network S1240; If the ratio of the number of pieces having a value equal to or larger than the set value (hereinafter, referred to as a classification success determination value) (hereinafter referred to as "classification success piece rate") is less than a preset ratio (hereinafter referred to as "permission level" A step of adjusting the number of pieces (S1260), and a step of verifying the accuracy of the classified result using the weight and the volume (S1270).

When an image to be analyzed is divided into a plurality of pieces using a grid, each piece of the grid may be a rectangle, or the aspect ratio of the rectangle may be different. In addition, each piece of the grid can be various polygons such as hexagonal or octagonal.

Adjusting the number of pieces (S1260) may include adjusting the number of pieces in a manner that increases the size of each piece of the grid, adjusting the number of pieces in a manner that reduces the size of each piece of the grid, And changing the horizontal-to-vertical ratio of each piece of the slice.

The step of verifying the accuracy may include displaying the result sorted in each piece of the grid, and displaying the sorted result may include displaying the results classified in each piece of the grid by color .

According to one embodiment of the present invention, the object to be analyzed may be a steel product to be subjected to an iron scrap for recycling, and the analysis result is an analysis result of the content of the iron scrap according to the steel scrap classification standard Lt; / RTI &gt;

If the analysis target is a steel product, the classification result may be obtained (S1240). Next, a step of calculating the content of the iron scrap according to the steel scrap classification standard of the steel product using the result sorted may be further included. And calculating the price of the steel product using the ratio of each steel scrap and the unit price of each steel scrap.

Since each step has been described in detail above, repeated explanation is omitted.

13 is a block diagram illustrating an iron scrap apparatus according to an embodiment of the present invention.

13, an iron scrap apparatus according to an embodiment of the present invention includes one or more processors 1310, a memory 1320 for loading a computer program executed by the processor 1310, A non-temporarily stored storage 1330, a display 1340 for displaying an analysis result of the image to be analyzed, a camera 1350 for capturing an image to be analyzed, a LiDAR sensor 1360 for measuring volume, 1370 &lt; / RTI &gt;

The processor 1310 controls the overall operation of each configuration of the computing device that may implement the method according to embodiments of the present invention. The processor 1310 may comprise a central processing unit (CPU), a microprocessor unit (MPU), a microcontroller unit (MCU), a graphics processing unit (GPU), or any type of processor well known in the art . In addition, processor 1310 may perform operations on at least one application or program to perform the method according to embodiments of the present invention. A computing device capable of implementing the iron scrap method according to embodiments of the present invention may include one or more processors.

Memory 1320 stores various data, instructions and / or information. The memory 1320 may load one or more computer programs from the storage 1330 to perform the iron scrap method according to embodiments of the present invention.

Storage 1330 may temporarily store one or more computer programs. The storage 1330 may be a nonvolatile memory such as ROM (Read Only Memory), EPROM (Erasable Programmable ROM), EEPROM (Electrically Erasable Programmable ROM), flash memory, etc., hard disk, removable disk, And any form of computer-readable recording medium known in the art. Storage 1330 stores a computer program for performing the iron scrap method according to embodiments of the present invention.

The camera 1350 can capture an image to be analyzed.

The display 1340 can display the analysis result of the analysis target image.

A computer program for performing the iron scrap method according to embodiments of the present invention may be loaded into memory 1330 and may include instructions to cause processor 1310 to perform the method according to embodiments of the present invention have.

Specifically, the computer program includes an instruction for dividing an image into a plurality of pieces using a grid, an instruction for analyzing each of the plurality of pieces by using an artificial neural network, a result obtained by classifying each of a plurality of pieces from an artificial neural network into a plurality of classification references The instruction to control the number of pieces, and the weight and the volume, when the ratio of the number of pieces with the accuracy of the classified result to the total number of pieces is equal to or greater than the predetermined value is less than the preset ratio, Lt; RTI ID = 0.0 &gt; accuracy. &Lt; / RTI &gt;

Instructions that control the number of pieces may include instructions to adjust the number of pieces in a manner that increases the size of each piece of the grid, instructions to adjust the number of pieces in a manner that reduces the size of each piece of the grid, To change the aspect ratio of the image.

According to one embodiment of the present invention, the object to be analyzed may be a steel product to be subjected to an iron scrap for recycling, and the analysis result is an analysis result of the content of the iron scrap according to the steel scrap classification standard Lt; / RTI &gt;

If the analysis target is a steel product, it may further include an instruction to calculate the iron scrap content ratio according to the iron scrap classification standard of the steel product using the result sorted after the instruction to obtain the classified result and the measured volume, And instructions for calculating the price of the steel product using instructions for calculating the scrap content ratio, followed by a ratio of each iron scrap content and a unit price of each scrap.

It will be understood by those skilled in the art that changes may be made in the sequence of steps or steps of one or more steps in the sequence of steps of the present specification without departing from the essential characteristics of the present invention, Various steps of each flowchart of this specification can be variously modified and modified, for example, by executing steps in parallel.

The methods according to the embodiments of the present invention described so far can be performed by the execution of a computer program embodied in computer readable code. The computer program may be transmitted from a first computing device to a second computing device via a network, such as the Internet, and installed in the second computing device, thereby enabling it to be used in the second computing device. The first computing device and the second computing device all include a server device, a physical server belonging to a server pool for cloud services, and a fixed computing device such as a desktop PC.

The computer program may be stored in a storage medium such as a magnetic storage medium (e.g., a ROM, a floppy disk, a hard disk, etc.), an optical reading medium (e.g. CD-ROM or DVD), a flash memory Or may be stored in the same non-transitory medium.

While the present invention has been described in connection with what is presently considered to be the most practical and preferred embodiment, it is to be understood that the invention is not limited to the disclosed embodiments, but, on the contrary, I can understand. It is therefore to be understood that the above-described embodiments are illustrative in all aspects and not restrictive.

Claims (22)

Measuring the weight of the steel product;
Measuring the volume of the steel product;
Photographing an image of the steel product;
Dividing the image into a plurality of pieces using a grid;
Analyzing each of the plurality of fragments using an artificial neural network;
Obtaining a result of classifying each of the plurality of fragments into a plurality of classification criteria from the artificial neural network;
Adjusting the number of pieces when the ratio of the number of pieces having an accuracy of the sorted result to the total number of pieces is more than a predetermined value is less than a predetermined ratio; And
And verifying the accuracy of the classification result using the weight and the volume.
Iron scrap method. The method according to claim 1,
Wherein each piece of the grid is rectangular,
Iron scrap method. 3. The method of claim 2,
Each piece of the grid having a different aspect ratio,
Iron scrap method. The method according to claim 1,
Wherein adjusting the number of pieces comprises:
And increasing the size of each piece of the grid.
Iron scrap method. The method according to claim 1,
Wherein adjusting the number of pieces comprises:
And reducing the size of each piece of the grid.
Iron scrap method. The method according to claim 1,
Wherein adjusting the number of pieces comprises:
And changing the aspect ratio of each piece of the grid.
Iron scrap method. The method according to claim 1,
After verifying the accuracy,
Further comprising displaying the grid on the image and displaying the sorted result on each piece of the grid.
Iron scrap method. 8. The method of claim 7,
The step of displaying the classified result includes:
And displaying the classified results in each piece of the grid in a color-coded manner.
Iron scrap method. The method according to claim 1,
The image is an image of a steel product loaded on a vehicle,
Wherein the plurality of classification criteria are classified into iron scrap classification criteria,
Iron scrap method. 10. The method of claim 9,
After obtaining the classified result,
Further comprising calculating the iron scrap content ratio according to the iron scrap classification standard of the steel product using the result of classifying each of the plurality of pieces and the volume,
Iron scrap method. 11. The method of claim 10,
After verifying the accuracy,
Further comprising calculating the price of the steel product using the weight, the content of each scrap, and the unit price of each scrap.
Iron scrap method. 11. The method of claim 10,
After verifying the accuracy,
Further comprising displaying the grid on the image and displaying the results of the iron scrap classification on each piece of the grid.
Iron scrap method. 13. The method of claim 12,
The step of displaying the classified result includes:
And displaying the results of the iron scrap classification on each piece of the grid in a color-coded manner.
Iron scrap method. Gauze devices for measuring the weight of steel products;
A LiDAR sensor for measuring the volume of the steel product;
A camera for photographing an image of the steel product;
A display for displaying an analysis result of the image;
One or more processors;
A memory for loading a computer program executed by the processor; And
And a storage for storing the computer program,
The computer program comprising:
An instruction to divide the image into a plurality of pieces using a grid;
Instructions for analyzing each of the plurality of fragments using an artificial neural network;
Instructions for classifying each of the plurality of fragments into a plurality of classification criteria from the artificial neural network;
Instructions for adjusting the number of pieces if the ratio of the number of pieces having an accuracy of the sorted result to the number of all pieces is equal to or greater than a predetermined value is less than a predetermined ratio; And
And an instruction to verify the accuracy of the classified result using the weight and the volume.
Scrap iron. 15. The method of claim 14,
After the instruction verifying the accuracy,
Further comprising instructions for displaying the image, the grid, and the classified result on the display,
Scrap iron. 16. The method of claim 15,
The instruction to be displayed on the display includes:
And displaying the sorted result on a color-coded display.
Scrap iron. 15. The method of claim 14,
Wherein the image is an image of a steel product loaded on a vehicle with the camera,
Wherein the plurality of classification criteria are classified into iron scrap classification criteria,
Scrap iron. 18. The method of claim 17,
After the instruction to obtain the classified result,
Further comprising instructions for calculating the iron scrap content ratio according to the iron scrap classification criteria of the steel product using the result of classifying each of the plurality of pieces and the volume,
Scrap iron. 19. The method of claim 18,
After the instruction verifying the accuracy,
Further comprising instructions for calculating a price of the steel product using the weight, the amount of each scrap, and the unit price of each scrap,
Scrap iron. 19. The method of claim 18,
After the instruction verifying the accuracy,
Further comprising instructions for displaying on the display classification results according to the image, the grid, and the iron scrap classification criteria.
Scrap iron. 21. The method of claim 20,
The instruction to be displayed on the display includes:
And displaying the sorting result on the display in a color-coded manner according to the iron scrap classification criteria.
Scrap iron. A computer program for causing a computer to perform the method of any one of claims 1 to 13.

Images