JP2007136430A - Method and apparatus for selecting dried sardine - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a method and an apparatus for selecting a plurality of kinds of object material to be selected such as dried sardine efficiently according to the kinds within a short time with no need of manual work or without damaging or breaking the object material to be selected at the time of selection. <P>SOLUTION: The apparatus comprises belt conveyers 12 to 17 on which the object material to be selected is arranged adjacently, a first sensor 21 for detecting existence of the object material to be selected, a second sensor 22 for detecting that the object material to be selected reaches the uppermost position, a separation apparatus 1 for separating the object material to be selected into one or more groups of the object material including a plurality of types, and an identification means for identifying the fish type of the dried sardine by inputting the images in an image frame as color images in a PC. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  This invention is for, for example, selecting “Iriko (boiled and dried)” consisting of a plurality of small fish of different fish types for each of the same fish species and removing different fish species and hybrids other than the selection target. The sorting material is not limited to “iriko”, but includes “seed”, “fresh fish”, “fruits and vegetables”, and the like.

  In the iriko sorting process, after landing, boiled and salted, the dried and semi-dried "Iriko" is classified by size through a sieve, and the sorted "Iriko" is placed on the sorting table in order of size. Therefore, the same type is selected manually. As is well known, Iriko has a plurality of fish species (see FIG. 4 (a)), such as red sea bream, junme sea bream, single-mouthed sea bream, sea bream fish and fire bream.

  However, the manual selection requires 5-7 years of experience because it requires skill, and the work takes time, and the work takes a long time and heavy labor. Moreover, since it has a long period of time to train successors, it leads to a shortage of workers. Furthermore, when the selection is not performed accurately and different fish species are mixed or small fish are mixed, the value of the product is significantly reduced. In addition, even in the case of manual work, the squirrel before sorting is usually piled up in layers and intertwined (bin state), so it is separated and taken out so as not to damage one by one. It is difficult to do.

  Therefore, the present invention can efficiently sort the materials to be sorted of multiple types such as “Iriko” for each material of the same type without using humans, in a short time, and damage the materials to be sorted at the time of sorting. It is an object of the present invention to provide a sorting method and a sorting device without breaking down.

In order to solve the above-mentioned problems, the method for sorting iriko etc. according to the present invention sorts a group of materials to be sorted in which a plurality of types of “iriko” etc. are mixed for the same type or excludes different types other than the selection target. A plurality of types of materials to be selected on the belt of the first belt conveyor among a plurality of belt conveyors that are inclined upward in the traveling direction and arranged in series so as to be transportable in order. The group is placed to start running the belt. When at least one material to be sorted reaches the uppermost position and falls on the belt of the second belt conveyor, the running of the first belt is stopped, and the second belt When the belt of the belt conveyor starts to run, and when at least one material to be sorted that has fallen on the belt reaches the uppermost position and falls on the belt of the third belt conveyor, the running of the belt is stopped, and the first belt Bertoco The operation of starting the traveling of the belt of the Bayer belt and the belt of the third belt conveyor is repeated until at least the third belt conveyor, preferably the fifth and subsequent belt conveyors, so that one or more materials to be selected are selected. When it is determined in the identification step described later that the material to be sorted and the material to be sorted separated into one or two or more are sorted in the identification step described later, while sorting to the sorting box corresponding to the type, When it is determined in the identification step described later that the materials to be separated separated into two or more are different types, the step of circulating to the separation step and the material to be separated separated in the separation step are The binarized image obtained by photographing the material to be sorted with a camera at a preset threshold value is subjected to frequency analysis processing using a two-dimensional fast Fourier transform. The template for each class composed of hierarchical neural network with a multilayer structure capable nonlinear identification, step or identify the type of the selected material from the binarized image by performing the matching processing using the non-linear template,
When there are two or more materials to be sorted separated in the separation step, an image obtained by photographing the material to be sorted with a camera is binarized with a preset threshold value, and 8-near contraction in the binarized image. After performing the processing, the labeling process is used to assign a label number to the part to be selected, and the center position in the label area is obtained, and one piece of the target material in the image is extracted using this center position information. A multi-layered hierarchical neural network capable of nonlinearly identifying a template for each type of material to be selected, and a step of performing frequency analysis using a two-dimensional fast Fourier transform (hereinafter also referred to as 2D FFT). And a step of identifying the type of the material to be sorted from the binarized image by performing a matching process using a non-linear template.

  According to the screening method of the present invention having the above-described configuration, first, after separating a group of materials to be sorted, such as a large number of groups of irises, into one to two and about three, the type of the material to be sorted is changed. Since the materials to be sorted are separated, the materials to be sorted easily separated into bins can be separated, and the number of materials to be classified to be identified can be reduced to the minimum, so that the identification can be performed easily and the identification time can be shortened. Further, even when the material to be sorted approaches after separation, the 8-neighbor contraction process is performed for extraction, so that the identification can be performed reliably. Furthermore, the input image data is converted into frequency components by two-dimensional fast Fourier transform, and the feature points of the image data can be easily grasped, so that information that is invariant to the displacement and rotation of the material to be sorted can be obtained. It is done. In addition, since the type of the material to be sorted is discriminated by the neural network, it is possible to correctly discriminate even a natural object, such as a natural object, which is easily affected by fluctuations in input information due to a certain shape, hue pattern fluctuation, and the like.

  In order to solve the above-described problems, the sorting apparatus for irises and the like according to the present invention (Claim 2) sorts a group of materials to be sorted in which a plurality of types of “irico” and the like are mixed for the same type or other than the selection target. A first sensor for inclining upward along the traveling direction, detecting the presence or absence of a material to be sorted placed on the belt, and starting the traveling of the belt A plurality of belt conveyors, each having a second sensor that detects whether the material to be sorted on the belt has reached the uppermost position and stops the belt travel or reverses the belt after stopping. A device for separating a plurality of types of materials to be sorted, in which materials to be sorted are arranged adjacent to each other so as to be transferable between belts, and a material separated into one or more materials. Sorting material is the same type When discriminated by the discriminating means described later, the sorting device sorts the sorting box corresponding to the type, and when the discriminating means discriminating the after-mentioned discriminates that the materials to be separated into two or more types are different, When there is one sorting means for circulating the group of materials to be sorted at the entrance and one material to be separated separated by the separating means, an image obtained by photographing the material to be sorted with a camera is set to 2 at a preset threshold value. A frequency analysis process is performed using two-dimensional Fast Fourier Transform, and a template for each type of material to be sorted is composed of a multi-layer hierarchical neural network capable of nonlinear identification, and a nonlinear template is used. Means for discriminating the type of material to be sorted from the binarized image by performing matching processing and two or more materials to be sorted separated in the separation step The binarized image obtained by photographing the material to be sorted by a camera with a preset threshold value, and after performing 8-neighbor contraction processing on the binarized image, label the material to be sorted using a labeling process. A number is assigned, a center position in the label area is obtained, one sort material in the image is extracted using the center position information, a frequency analysis process means using two-dimensional fast Fourier transform, A template for each type of material to be sorted is composed of a hierarchical neural network having a multi-layer structure capable of nonlinear identification, and matching processing using the nonlinear template is performed to obtain the material to be sorted from the binarized image. And a means for identifying the type.

  According to the sorting apparatus such as Iriko having the above-described configuration, the sorting method according to claim 1 can be performed, and the same sorting method as the above-described working effect can be obtained. Since the conveyors are separated in combination, the structure of the apparatus is relatively simple, and even those that are easily damaged during separation can be reliably separated with minimal damage.

  Further, in order to solve the above-mentioned problems, a sorting apparatus such as an iriko according to the present invention (Claim 3) includes a plurality of belt conveyors inclined upward and arranged adjacently so that materials to be sorted can be sequentially transferred, A first sensor that senses the presence or absence of a material to be sorted on the lower part of the belt of the belt conveyor and instructs the start of rotation of the belt when the material to be sorted is detected, and the material to be sorted on the belt of each belt conveyor reaches the uppermost position. And a second sensor for instructing to stop the rotation of the belt when it is detected, and separating a plurality of types of materials to be sorted into one or two or more materials to be sorted in a confused state or intertwined in bulk An apparatus, a camera for photographing a material to be sorted separated by the separation device, and a case where the material to be sorted is 1 in the image frame photographed by the camera; When all the materials to be sorted are determined to be of the same type by the discriminating means, there are 2 means for sorting the materials to be sorted into corresponding types of sorting boxes, and 2 materials to be sorted in the image frame photographed by the camera. When it is determined that the material to be sorted is of a different type by the following determination means, the recirculation means for returning to the separation device, the image in the image frame is captured as a color image in a personal computer, and the threshold value is set. Use binarization to convert to a monochrome image, then perform 8-neighbor contraction processing, perform labeling, then find the center position, reduce it using the billnea method, cut out, cut out The binarized image is binarized using a threshold value that can distinguish the material to be sorted from the background, the background is filled with black, and the monochrome image is converted using a two-dimensional fast Fourier transform. Calculate the wave number component, block the frequency component concentrically, normalize the sum of the blocks with the same number, and calculate the value to be input to the neural network. Means for inputting to a neural network and identifying the type of the material to be sorted.

  According to the sorting apparatus such as the litter having the above-described configuration, there are substantially the same effects as the sorting apparatus according to the second aspect.

The sorting method for the sardine according to the present invention and the apparatus have the following excellent effects. That is,
-Conventionally, the manual sorting work such as sardines can be performed efficiently and reliably in a short time, labor saving and high accuracy can be achieved, and mixed fish and different fish species can be mixed. It is difficult to improve product value.

  -Even when the material to be sorted approaches after separation, the extraction is performed by performing the 8-neighbor contraction process, so that the identification can be performed reliably.

  ・ By converting the input image data into frequency components by two-dimensional fast Fourier transform so that the feature points of the image data can be easily grasped, it is possible to obtain information that is invariant to displacement and rotation of the material to be selected. In addition, since the type of material to be sorted is determined by a neural network, it can be correctly identified even if it is a natural object such as a natural object and is susceptible to fluctuations in input information due to variations in shape, shade pattern, etc. .

  ・ Since a plurality of belt conveyors are combined and separated, the structure of the difference is relatively simple, and even those that are easily damaged during separation can be reliably separated with minimal damage.

  DESCRIPTION OF EMBODIMENTS Hereinafter, embodiments of a sorting apparatus and a sorting method, such as a sardine, according to the present invention will be described with reference to the drawings.

  FIG. 1 is a flowchart showing an embodiment of a sorting method when the material to be sorted is Iriko. FIG. 2 (a) is a side view showing an iroko separating device as one component of the sorting device, and FIG. 2 (b) is an enlarged side view showing one belt conveyor.

  In this embodiment, as shown in FIG. 4 (a), five kinds of sardines are selected as a selection target, such as true sardines, junme sushi, single-mouthed swords, sea bream fish and fire bream.

Therefore, as shown in FIG.
1) After visually selecting the squid to be selected for each fish species (for example, 10) for each fish species, the fish species are photographed with a USB-connected digital camera for each fish species of the same fish species. A pattern is input to a neural network (hereinafter referred to as NN) and registered for learning.

  The sorting apparatus 1 according to the present embodiment includes an identification unit 10, and the identification unit 10 is configured by using NN as a template for discriminating fish species based on the overall shape, hue, and side line pattern of the iris, and matching using a nonlinear template. Process. The structure of the NN is assumed to be a three-layer structure having a minimum structure capable of nonlinear identification. Since the shape of iriko is not stable, if the color and the side line pattern finally match, it is determined that the fish species are the same. Regarding the NN configuration method, two types of target units (for example, a unit that reacts only to authenticity) and non-target units (for example, a unit that responds to other than authenticity) are set for the output pattern in the output layer. It is assumed that the final judgment is performed based on the magnitude relationship of values. Specifically, if the reaction value of the unit that reacts only to the true snapper is larger than the reaction value of the unit that reacts to anything other than the true snapper, it is determined that the fish species to be selected is the true snapper.

  2) The sardine group S to be separated / sorted is put into the separating apparatus 1 (FIG. 2) in the bin state.

  3) Operate the separation device 1 to separate the iriko into one or more animals.

  4-1) Take a picture of the swords separated into two or more animals with the digital camera 3, and extract each one of them from the photographed image.

  4-2) Identify each fish species by NN as described later.

  5) As a result of the identification, one or two or more squirrels made of the same fish species are distributed to a selection (by type) box corresponding to the fish species.

  6) As a result of the identification, two or more sardines made of different fish species are returned to the syllabus group S before separation in 2) above, or thrown back into the entrance of the separation device 1 to perform so-called reflux.

  At the time of sorting, the extraction process is required to identify individual fish species from the “Iriko” group in the bin state. Moreover, in order to extract, it is necessary to isolate | separate a sardine. The separating apparatus 1 developed for this purpose is composed of one inclined belt conveyor 12 with a conveying rib and five separating inclined belt conveyors 13 to 17, as shown in FIG. . The driving of the belt conveyors 12 to 17 is made independent, and the forward rotation driving, reverse rotation driving, and stop control are performed by a microcomputer board. The conveyor belt conveyor 12 is provided with a hopper 18 for feeding the iroko group S. When the irrig group S is introduced into the hopper 18 and the endless ribbed belt 12a starts to travel, the iriko group S moves upward. Carried to.

  Each of the separation belt conveyors 13 to 17 is inclined slightly upward in the traveling direction of the belts 13a to 17a, and can transfer the object (iriko) S to the belt of the next adjacent belt conveyor. In this way, they are arranged in parallel (columns) at regular intervals. Each of the separation belt conveyors 13 to 17 has a structure in which the endless belts 13a to 17a are caused to travel together with the conveyor belt conveyor 12 by a DC motor 19 via a reduction gear 20, and are respectively arranged at an intermediate position and an uppermost position in the belt traveling direction. Near-infrared sensors 21 and 22 are attached. The first sensor 21 at the intermediate position is directed to the lower end on the endless belt. When the first sensor 21 is detected on the endless belt, the DC motor 19 is driven to start running (rotation) of the endless belt 13a. On the other hand, when the second sensor 22 at the uppermost position senses an urine, it stops the travel of the endless belt 13a. For example, it is assumed that a plurality of dogs have dropped from the uppermost position of the belt 12a of the conveyor belt conveyor 12 and have moved to the lower end of the next conveyor belt 13a. The transferred number of irrigators move upward as the belt 13a travels. When the belt 13a is moved to the uppermost position, it is detected by the second sensor 22 and the running of the belt 13a is stopped. One to two or three dogs fall on the lower end of the belt 14a of the next conveyor belt 14. Hereinafter, when the belts of the conveyor belts 14, 15, and 16 are rotated in order, and the sword falls on the belt 17a of the fifth conveyor belt 17, the belt 17a travels a predetermined distance, and the sword S moves in the traveling direction of the belt 17a. Move to almost the middle position.

  Since the endless belts 12a to 17a are made of an elastic body such as rubber, they are not damaged even if they are dropped, and the spine S is prevented from rotating during belt running. When the belt conveyors 12 to 16 adjacent to each other stop rotating on one belt conveyor (for example, 13), the microcomputer board controls the other belt conveyor (for example, 14) to start rotating the belt. Therefore, in the five belt conveyors of this example, the belts 12a to 16a rotate every other unit. In this way, a large number of sardine groups S are placed on the belt 17a of the fifth belt conveyor 17 in a state of being separated into about 1 to 2.3.

  Above the fifth belt conveyor 17, the luminaire 2 and the digital camera 3 are arranged with their lenses facing approximately the middle position in the running direction of the belt 17 a of the belt conveyor 17. In the fifth belt conveyor 17, the rotation of the belt 17a is started when the presence of the sensor 21 is detected by the sensor 21, but the rotation of the belt 17a is stopped when the slit is in the intermediate position. Therefore, the sword on the belt 17a is photographed by the digital camera 3.

  This image is sent to a personal computer (hereinafter also referred to as a PC), binarized by a threshold value, and converted into a white image with a black background. In order to convert the converted image data into a numerical value suitable for the input of a neural network (hereinafter referred to as NN), a frequency analysis process is performed using a two-dimensional fast Fourier transform (hereinafter also referred to as 2DFFT). Block dimension data concentrically. The blocks having the same numbers shown in FIG. 3F are averaged and normalized to obtain a slab value input to each unit (hereinafter also referred to as a cell or a neuro element) of the input layer. Since the effective area of the component obtained from 2DFFT is 64 × 64, the slab value is 64.

  In this example, the arithmetic unit having a three-layer structure includes three layers of an input layer (64 neuro elements), a hidden layer (35 neuro elements), and an output layer (5 neuro elements), and the input layer is obtained from 2DFFT. The slab value calculated for each effective region of the component is input to the corresponding neuro element. In this example, the hidden layer is composed of 35 neuro elements, and plays a role of propagating the input layer information and transmitting it to the output layer. As the number of hidden layers increases, it is possible to perform the separation operation on each pattern without changing the slab value of the input layer. The output layer is provided with five neuro elements so as to correspond one-to-one for each type of “Iriko” to be identified. Then, the output neuro element values based on the weighting factor between the neuro elements completed by learning are calculated for several output neuro elements.

  Although partially repeated, a method for identifying (determining) this fish species using NN will be described in detail below.

  1) In this example, an RGB color image captured by a USB-connected digital camera is taken into a PC (FIG. 3A).

  2) Binarization is performed using a threshold value and converted to a monochrome image (FIG. 3B).

(R × 0.33 + G × 0.33 + B × 0.33) / 3 = X
3) Perform 8-shrinkage treatment (shrinkage treatment in the left, right, up, down, and slant neighborhoods), separate the material to be sorted (Iriko) in the overlapping state, and perform labeling. Subsequently, the center position is obtained and reduced to 128 × 128 (2 to the power of 2 × 2 to the power of n) using the billier method (FIG. 3C).

  4) The cut-out image is binarized using a threshold value that can distinguish between the background and the background, and the background is filled with black (0 × 00) (FIG. 3D).

5) Using the 2D FFT according to the following formula, the monochrome image (128 ×
128 pixel) frequency components are calculated (FIG. 3E).

  6) The frequency components arranged as shown in FIG. 3 (f) are concentrically blocked, and the values input to NN are normalized by adding and averaging the blocks having the same number. Ask.

For example, a value obtained by adding only the points “1” in FIG. 3F and taking the average is one of the input elements of the NN. Σx (i) / 12 (i = 1,2,… 12)
Other input elements have the same calculation format, Σx (i) / M (i = 1,2, ... M) where M is the number of the same number of concentric circles, that is, 0, 1, 2, 3, 4, 5 .......... 62.63.64
The input value thus created is input to the NN, and the fish species are identified (FIG. 3 (g)).

Here, the basic configuration of the identification unit 10 will be described in more detail. FIG. 5 is an explanatory diagram showing the basic configuration of the identification means 10. Each neuro template in FIG. 5 individually identifies this fish species. In this identification process, this image is processed by 2DFFT which is preprocessing, and a slab value which is an input value to the NN is calculated. In this case, 64 values are input. Next, this value is input to each neuro template, and NN forward calculation is performed. Individual neurotemplates only determine whether individual fish species are identified or otherwise. For example, when an image of authenticity is input, if the first neurotemplate i is for authenticity, the unit (cell) of the output layer of this neurotemplate i
There are two units, a unit indicating authenticity (target pattern) and a unit indicating non-authentic (non-target pattern).

  Therefore, since the correct image of the correct authenticity is inputted, the unit of the output layer of this neuro template i shows the following relationship.

Response value of output unit (unit indicating authenticity)> Response value of output unit (unit indicating non-authenticity) ............ (1)
Similarly for other neuro templates ii to v, Junmoku, Kataguchi, and Kobiko
It is a setting that uniquely identifies each of them, such as firefly sauce. However, the same authentic image is input to other neuro templates, and it is confirmed whether the following relationship is satisfied.
That is,
Response value of the output unit (unit indicating the moisture) <Response value of the output unit (unit indicating the other) …… (2)
Response value of the output unit (unit indicating one port) <Response value of the output unit (unit indicating other) …… (3)
Response value of the output unit (unit indicating the baby) <Response value of the output unit (unit indicating the other) …… (4)
Response value of output unit (unit indicating a crack) <Response value of output unit (unit indicating other) …… (5)
In other words, it is confirmed that it is authentic with a true neuro template i, and it is a neuro template ii-v that uniquely identifies Junmoku, Kataguchi, Kibago, and hototare, not Junmoku, Make sure that you are neither a single-mouthed moth, nor a kid, nor a firefly, and make a final decision.
High-precision identification can be realized by template matching by the NN. However, as shown in FIG. 6, the conventional sorting apparatus 1 is a method in which units corresponding to the respective fish species are set as they are in the output layer of the conventional NN and the fish species are determined by only one NN calculation. The program system can be selected when the sorting apparatus 1 is activated.

  In this way, the fish species on the conveyor belt 17a are identified. However, if there is only one irico and two or more fish species are the same, the conveyor belt 17a is rotated to the uppermost position. After that, it is sorted by a sorting device (not shown) into a sorting box (not shown) corresponding to the fish type.

  On the other hand, when there are two or more squirrels and the fish species are different, they are returned to the separation device 1 again by a sorting device (not shown) and the separation work is repeated. When the separation operation is repeated, since most squid are finally separated one by one, they are distributed to the selection box corresponding to the fish species.

  By the way, with regard to the picking apparatus according to the embodiment of the present invention, all the fish species to be sorted are photographed in advance with a digital camera, and the fish species data (fish species pattern) based on the image data is NN. Is registered for learning. Then, among the image data, data relating to characteristic patterns (patterns) that are important in identifying fish species such as the overall shape, color, and side line pattern, etc. of the image data are obtained by using a two-dimensional fast Fourier transform as shown in FIG. The frequency analysis process is performed, and the input pattern (slab value) is input and stored in the NN input layer. Then, the separation apparatus 1 normally separates one to two or three squids and the image data photographed by the digital camera 3 is similarly converted along the operation flow of FIG. Is transmitted and processed from the input layer to the hidden layer to the output layer, and an output value (output pattern) is output from the output layer. This output value is based on the weighting coefficient obtained by the learning so far, and in the case of this example, what the fish species are is output.

  On the other hand, when a correct output value is given to the fish species pattern in NN, it is transmitted and processed in the order of output layer → hidden layer → input layer, and the weighting coefficient between each layer is learned. Learning the weighting factor is to change and converge the strength of the coupling between the neuro elements of the input neuro element, hidden neuro element, and output neuro element so that the difference between the actual output value and the correct output value is reduced. .

Specifically, with regard to NN learning, the configuration of the NN uses a hierarchical three-layer NN, and adopts an improved error back propagation method as a learning method. This improved error reverse carrying method is a learning algorithm expressed by the following equation (1).

  Here, W is a weight, t is the number of learnings, δ is a generalization error, O is an output value of a neural element (cell / unit), η is a learning constant, α is an inertia constant, and β is a vibration constant.

  Then, the fish species pattern as the target material to be selected is compared with a plurality of types of fish species patterns that have been registered in advance. By comparing with the fish species pattern, the fish species can be identified more accurately.

Although the example of the sorting apparatus according to the present invention has been described above with reference to an embodiment, it can be implemented as follows. That is,
・ It can be used not only to select sardine groups consisting of multiple fish species for each same species, but also to remove only sparse fish from sardine groups mixed with small fish.

  -Sorting object is not limited to fish, but can be applied to a device that sorts or monitors only one side or the front side of a material to be sorted.

It is a flowchart which shows the Example of the sorting method in case a to-be-sorted material is Iriko. FIG. 2 (a) is a side view showing an iroko separating device as one component of the sorting device, and FIG. 2 (b) is an enlarged side view showing one belt conveyor. (A)-(c) is explanatory drawing which shows a sorting operation | movement flow in order. (D)-(g) is explanatory drawing which shows in order the flow of selection operation | movement of an iris. FIG. 4 (a) is a front view showing the fish species (five types), and FIG. 4 (b) is a front view showing a state of sorting for each same type. It is explanatory drawing which shows the basic composition of a selection means (discrimination means). It is explanatory drawing which shows the identification structure in the conventional NN.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Separation apparatus 2 Lighting fixture 3 Digital camera 10 Identification means 12 Inclination belt conveyors 13 to 17 for conveyance Inclination belt conveyors 12a to 17a for separation Conveyor belt 18 Hopper 19 DC motor 20 Reduction gears 21 and 22 Sensor

Claims (3)

A sorting method for sorting a group of materials to be sorted with multiple types of iriko, etc., for each same type or excluding different types other than the selection target,
Among the multiple belt conveyors that are tilted upward in the running direction and arranged in series so that they can be conveyed in sequence, a plurality of types of materials to be sorted are placed on the belt of the first belt conveyor, and belt running starts. When at least one material to be sorted reaches the uppermost position and falls onto the belt of the second belt conveyor, the traveling of the first belt is stopped, the belt of the second belt conveyor is started, When at least one material to be sorted that has fallen on the belt reaches the uppermost position and falls on the belt of the third belt conveyor, the traveling of the belt is stopped, and the belts of the first belt conveyor and the third belt conveyor are stopped. The operation of starting the running of the belt is repeated until at least the third belt conveyor, preferably the fifth and subsequent belt conveyors, so that one or two to-be-sorted materials are collected. And separating the to be sorted material,
When the materials to be sorted separated into one or more are determined to be of the same type in the identification step described later, the materials to be sorted are sorted into the sorting boxes corresponding to the types, while the materials to be sorted separated into two or more Are determined to be different types in the identification step described later, the step of refluxing to the separation step,
If there is one material to be sorted separated in the separation step, the image obtained by photographing the material to be sorted with a camera is binarized at a preset threshold value and subjected to frequency analysis processing using a two-dimensional fast Fourier transform. The material to be sorted is composed of a multi-layered hierarchical neural network capable of nonlinear identification for each type of material to be sorted, and the binarized image is sorted by performing matching processing using the nonlinear template. The process of identifying the type of material, or
When there are two or more materials to be separated separated in the separation step, an image obtained by photographing the material to be sorted with a camera is binarized with a preset threshold value, and 8-near contraction in the binarized image is performed. After performing the processing, the labeling process is used to assign a label number to the portion to be selected, and the center position in the label area is obtained, and one piece of the selection material in the image is extracted using this center position information. Processes, frequency analysis processing using two-dimensional fast Fourier transform, and materials to be sorted are composed of multi-layered hierarchical neural networks that can identify nonlinearly for each type of materials to be sorted. And a step of identifying the type of the material to be sorted from the binarized image by performing matching processing. A sorting device for sorting a group of materials to be sorted in which a plurality of types of sardines are mixed for the same type or excluding different types other than the selection target,
A first sensor that tilts upward along the running direction, detects the presence or absence of a material to be sorted placed on the belt, and starts running of the belt, and that the material to be sorted on the belt has reached the uppermost position. A plurality of belt conveyors each provided with a second sensor that detects and stops the belt traveling or reverses after the belt is stopped are arranged adjacent to each other so that a material to be sorted can be transferred between adjacent conveyor belts. An apparatus for separating a plurality of types of materials to be sorted into one or more materials to be sorted,
When the material to be sorted separated into one or two or more is identified by the identification means described later as being of the same type, the materials to be sorted are sorted into the sorting box corresponding to the type, while being separated into two or more. When the discriminating means to be described later determines that the different types are different, the sorting means for circulating the group of materials to be sorted at the entrance of the separation device,
When the material to be sorted separated by the separating means is one, an image obtained by photographing the material to be sorted with a camera is binarized at a preset threshold value, and a frequency analysis process using a two-dimensional fast Fourier transform The material to be sorted is composed of a multi-layered hierarchical neural network capable of nonlinear identification, and a template for each type of material to be sorted is subjected to matching processing using the nonlinear template. Means for identifying the type of sorter; and
When there are two or more materials to be sorted separated in the separation step, an image obtained by photographing the material to be sorted with a camera is binarized with a preset threshold value, and 8-near contraction in the binarized image. After performing the processing, the labeling process is used to assign a label number to the part to be selected, and the center position in the label area is obtained, and one piece of the target material in the image is extracted using this center position information. Means for frequency analysis processing using two-dimensional fast Fourier transform, and the material to be sorted is composed of a multi-layered hierarchical neural network capable of nonlinear identification for each type of material to be sorted. And a means for discriminating the type of the material to be sorted from the binarized image by performing matching processing. A plurality of belt conveyors that are inclined upward and adjacent to each other so that the materials to be sorted can be transferred to each other, and the presence or absence of the materials to be sorted are detected on the belt lower part of each belt conveyor. A first sensor for instructing the start and a second sensor for instructing to stop the rotation of the belt when it senses that the material to be sorted on the belt of each belt conveyor has reached the uppermost position. A separation device for separating a plurality of types of materials to be sorted into one or more materials to be sorted,
A camera for photographing the material to be sorted separated by the separation device;
When the material to be sorted is 1 in the image frame photographed by the camera and when there are 2 or more materials to be sorted and all the materials to be sorted are determined to be the same type by the following discriminating means Means for distributing the materials to the corresponding types of sorting boxes;
Recirculation means for returning to the separation device when the material to be sorted is determined to be two or more in the image frame photographed by the camera, and the material to be sorted is determined to be of a different type by the following discrimination means;
The image in the image frame is taken into a personal computer as a color image, binarized using a threshold value, converted into a monochrome image, subjected to 8-neighbor contraction processing, labeling, and then the center position is determined. Obtained, reduced using the bilnear method, cut out, binarized the cut out image using a threshold that can distinguish the material to be sorted from the background, painted the background in black, and 2D fast Fourier transform Is used to calculate the frequency component of the monochrome image, block the frequency component concentrically, add the same number of blocks, and normalize the value to be input to the neural network. And an identification means for identifying the type of the material to be sorted by calculating and inputting the input value to the neural network. Sorting apparatus.