JP4649628B1 - Whole surface image inspection device for spheroid and spherical body - Google Patents

Abstract

【Task】
The classification of the equal class by image processing uses the appearance of the selection object as the selection criterion, so it is necessary to inspect the entire surface of the selection object, but the selection object is placed on a conveyor that forms part of the transport unit. Therefore, there has been a problem in that imaging of the conveyance unit and the contact portion to be selected cannot be performed. Further, if the selection targets are in a piled state, image processing cannot be performed properly, and the screens must be separated. There has been a demand for labor saving by automating sorting work that has been done manually.
[Solution]
By combining a plurality of independently driven V-shaped belt conveyors and a holding mechanism, it is possible to deal with multiple types of sorting objects with various shapes and dimensions, and promote separation of piled sorting objects. Thus, the sorting operation is separated into one by the imaging unit, and the sorting operation is automated by full-surface image processing using a neuro network or the like in the identification unit.
[Selection] Figure 1

Description

  The present invention relates to an apparatus for imaging an inspection object carried by a conveying means such as a conveyor with a plurality of cameras and performing an appearance inspection of the inspection object based on the captured images.

  At present, in the fruits and vegetables market, the equalization of fruits and vegetables is regarded as important for the purpose of improving the product value. For this reason, efforts are being made to equalize the equal class in each farmhouse and region for the purpose of branding fruit and vegetable products. On the other hand, most of the sorting machines used for sorting fruits and vegetables use weight as the sorting standard, and the sorting work such as shape and scratches is visually compared with the same class standard table by multiple people over a long period of time. , Matching and sorting.

  In the isoclass standard table, the weight selection criteria are clearly defined by the number of grams. However, since the selection criteria for shapes, scratches, etc. are comparisons with photographs showing respective defects, no clear selection criteria are shown. For this reason, it is difficult to make the sorting standards uniform due to individual differences in the sorting standards, and a reduction in the commercial value is a problem. Furthermore, since the manual sorting operation requires checking the entire surface of the fruits and vegetables, a long work is required, which is a burden on the worker.

  Development of an apparatus that automates sorting operations of equal grades of agricultural products by partially using a belt conveyor as a conveying means and image processing as an identification means has been performed (for example, Patent Document 1).

  As a concrete example, after picking up the front surface of the fruits and vegetables, it is sandwiched with a roller, the fruits and vegetables are inverted, and the contact portion is imaged (for example, Patent Document 2), or when the fruits and vegetables move between the conveyors In addition, the development of a technique for imaging the contact portion (for example, Patent Document 3) and a technique for lifting the selection object using the suction pad and imaging the contact portion between the transport section and the fruits and vegetables (for example, Patent Document 4) are attempted. It has been.

  In the above prior art, fruits and vegetables have been picked up as the selection target, but besides the fruits and vegetables as the selection target, for example, there are processed fishery products, industrial products, etc., and there are jewelry such as pearls, Selection is automated by image processing (for example, Patent Document 5 and Non-Patent Document 1).

Japanese Patent Application Laid-Open No. 11-13239 JP 2000-084495 A JP 2000-237696 A JP 2001-004342 A JP-A-5-35848

“Development of a selection part of a scouring selection system using a neural network”, Proceedings of the 49th Research Presentation of the Institute of Systems, Control and Information Engineers, pp515-516, 2005

  The methods shown in the above-mentioned patent documents and non-patent documents have a problem that the methods that can be used are limited depending on the size and shape of the selection target (for example, fruits and vegetables), and thus cannot be applied to a plurality of types of selection targets.

  Furthermore, the classification of the equal class by image processing uses the size, scratches, and shape of the sorting target as a sorting criterion, and scratches and color irregularities often appear on a part of the surface. Since it is necessary to inspect the entire surface, the object to be sorted is placed on the conveyor that forms part of the transport means, so the contact portion of the conveyor and the object to be sorted (the bottom of the object to be sorted) cannot be imaged. Was a problem.

  As a means for solving this problem, there has been proposed a technique (Patent Document 4) for picking up an object to be sorted using a suction pad and imaging a contact portion between the conveyor and the object to be sorted. Since the contact portion with the conveyor is imaged after imaging, there is a problem that it takes time to image and identify the entire selection target.

  In order to improve the efficiency of the sorting work, a large amount of sorting objects are put into the transport means and transported to the identification means. If a large number of sorting objects are piled up, the identification and sorting process is appropriately performed. Can not do. In order to perform appropriate image processing, each of the selection targets must be separated.

  Furthermore, since the sorting work is currently performed manually, it is difficult to make the sorting standard uniform due to individual differences among workers. Moreover, since the worker is forced to work for a long time, automation of the sorting operation is required.

  In order to solve the above-mentioned problem, the present invention separates the selection targets put in a pile on the transport means, and performs whole-surface imaging of each selection target including the contact portion with the transport means, By automating the sorting operation that has been performed manually, it is intended to save labor and make the sorting standard uniform.

In order to solve the above-mentioned problems, a spheroid and spherical body full-surface image inspection apparatus according to the present invention holds a spheroid or spherical body sorting object, a conveying unit that conveys the spheroid or spherical object sorting object, and holds the sorting object following the conveying unit A spheroid and a spherical body including a holding mechanism that performs imaging, an imaging unit having a plurality of cameras that capture the selection target, and an identification unit that identifies and determines the plurality of images of the selection target captured by the imaging unit A full-scale image inspection apparatus,
The transport unit includes a plurality of V-shaped conveyors obtained by combining two flat belt conveyors so that the cross section in the belt traveling direction is V-shaped, and each of the plurality of V-shaped conveyors has an independent drive mechanism. Each of the V-shaped conveyors is installed with an inclination so as to convey the sorting object upward, from the upstream to the downstream of the conveying direction, and the holding mechanism is It has a plurality of string-like holding parts for conveying and holding the sorting object, is driven independently from the V-shaped conveyor, and the starting end of the string-like holding part is the lowest V of the conveying part It is arranged at the end of the mold conveyor. (Claim 1)

  According to this configuration, since the V-shaped conveyor is used for the transport unit, it is easy to separate the selection targets one by one near the top of the V-shape in the process of transporting the selection targets. Are arranged adjacent to each other so that they can be sequentially transferred, so that the sorting object is separated when transferring from the previous stage to the next V-shaped conveyor. Moreover, it is preferable that the number of V-shaped conveyors constituting the transport unit is three or more so that selection targets are separated and transported to the holding mechanism one by one. Note that the optimum number of V-shaped conveyors is determined by specifications (shape dimensions, etc.) to be selected, specifications as an inspection apparatus, and the like, and varies depending on the case by case. Generally, if the number of V-shaped conveyors increases, the separation of sorting objects proceeds, but the cost increases accordingly. In addition, the length and width of each V-shaped conveyor are determined to be optimum values depending on the specifications (shape dimensions, etc.) to be selected and the specifications of the entire surface image inspection apparatus.

  Each V-shaped conveyor is shifted down in order, specifically, the starting end of the V-shaped conveyor of the next stage is arranged with the stage lowered from the terminal end of the preceding V-shaped conveyor. Each V-shaped conveyor is sequentially transported from the upstream to the downstream while being transferred while descending. Separation of the sorting object proceeds when the V-shaped conveyor is transferred. In addition, upstream means the start end direction (sorting target input portion side) in the belt traveling direction, and downstream means the end direction (holding mechanism side) of the transport portion.

  Since each V-shaped conveyor has an independent drive mechanism, the conveyance speed can be adjusted for each V-shaped conveyor. If there is a difference in speed, there may be an advantageous effect on the separation of the objects to be sorted.

  Since the plurality of V-shaped conveyors are installed with an inclination with respect to the horizontal plane, the sorting object is conveyed upward from the upstream to the downstream.

  Since the holding mechanism is driven independently of the V-shaped conveyor, the conveyance speed can be set to a speed different from that of the V-shaped conveyor.

  A string, a thread, a rope, or the like can be used for the string-like holding unit, and the diameter thereof can be changed according to a selection target. If the selection target is heavy, the thicker one is preferable, but the thinner one is generally convenient for imaging. As will be described later, if a light-transmitting material (for example, a transparent thread) is used for holding the string-like shape, it is convenient because it does not hinder the imaging of the selection target. The string-like holding part is composed of at least two for the convenience of holding the selection target at the upper part thereof. You may provide a 3rd string-like holding part in a position lower than two string-like holding parts in the middle of two string-like holding parts. In this way, the selection target is supported at three points and the posture is stabilized, and the selection target can be held even when the selection target is smaller than the distance between the two string-like holding portions. Since the selection target is held by the two or three string-like holding units, the string-like holding unit is unlikely to obstruct the image pickup, and the image can be easily picked up from below the selection target.

  In the identification determination in the identification unit, the selection target is identified in a predetermined category, and determination of pass / fail is performed based on the identification result.

  As the selection target, a spheroid or spherical body is mainly preferred, and it may be a natural product such as fruits and vegetables, agricultural products, and marine products, and may be industrial products or jewelry such as pearls. There may be. Since the spheroid or spherical body is likely to roll under the influence of gravity, in the case of the V-shaped conveyor inclined upward according to the present invention, it is possible to promote separation of the selection objects.

  In the spheroid and spherical whole-surface image inspection apparatus according to the present invention, it is preferable that the string-like holding portion has light transmittance. (Claim 2)

  According to this configuration, if the string-like holding part is light-transmitting, the string-like holding part does not easily interfere with imaging, and it is possible to easily pick up an image from below the selection target. More preferably, by using a relatively thin transparent thread or teg for the string-like holding part, the determination accuracy is further increased without interfering with imaging.

  In the spheroid and spherical whole-surface image inspection apparatus according to the present invention, the number of the cameras is six, and the plurality of images are picked up from six different directions to be selected. (Claim 3)

  According to this configuration, a video obtained by imaging the selection target from six different directions is obtained as an image to be identified. If six cameras are installed so as to constitute an orthogonal coordinate system, a six-sided view of the selection object can be obtained.

  In imaging, the imaging environment is preferably illuminated by indirect illumination. Direct illumination may cause unwanted specular reflection such as halation in the selection target, and may cause a determination error in determining scratches.

  In the spheroid and spherical whole-surface image inspection apparatus according to the present invention, an imaging position detection unit that detects that the selection target has reached a predetermined position of the holding mechanism, and a command from the imaging position detection unit And a synchronous imaging mechanism for simultaneously imaging the selection object by the camera. (Claim 4)

  According to this configuration, when the selection target reaches a predetermined position suitable for imaging, the selection target is detected, the selection target is stopped, and synchronous imaging is performed by a plurality of cameras. Although the synchronous image may be taken while the selection target is moving, if the image is taken while moving, the imaging processing speed is increased and the operating rate of the entire surface image inspection apparatus is improved, but the video is slightly blurred. Here, the synchronous imaging means that the selection target is imaged with the camera at the same timing.

  As the imaging position detection means, the selection target in the holding mechanism is monitored, and compared with an image when there is no selection target, it is determined that the selection target has arrived at the imaging position. You may detect temporally using a timer etc. irrespective of a comparison with an image. Alternatively, the determination may be made based on an image captured using a color filter for hue to be selected.

  In the spheroid and spherical whole-surface image inspection apparatus according to the present invention, an optical sensor is attached to a starting end portion of each of the V-shaped conveyors excluding at least the uppermost stage and a terminal end portion of the lowermost V-shaped conveyor, and the optical When the sensor detects that the object to be sorted is transported from the previous stage to the next V-shaped conveyor or from the lowermost V-shaped conveyor to the holding mechanism, the previous or lowermost V-shaped conveyor is transported. Or when the optical sensor detects that the sorting object has been transported from the previous stage to the next V-shaped conveyor or from the lowermost V-shaped conveyor to the holding mechanism. , Means for reversely feeding the V-shaped conveyor at the preceding stage or the lowermost stage for a certain period of time. (Claim 5)

  According to this configuration, it can be detected by the optical sensor whether or not the sorting object has been transferred from the preceding V-shaped conveyor to the succeeding V-shaped conveyor. In addition, the optical sensors are arranged on both sides in the belt width direction near the starting end of each V-shaped conveyor and near the terminal end of the lowermost V-shaped conveyor. Note that the optical sensors may be directly attached to them, or may be separately arranged.

  Then, when it is detected by the optical sensor that the selection target has been transferred to the next V-shaped conveyor, the selection target can be separated more efficiently by stopping or reversely feeding the previous V-shaped conveyor. In addition, when the selection target is the spheroid or spherical body described above, if the V-shaped conveyor in the previous stage is stopped, the selection target can be separated efficiently by rolling backward due to gravity, and the selection target is a square that is difficult to roll. In some cases, separation of the selection target can be promoted by reversing the preceding V-shaped conveyor.

  In the spheroid and spherical whole-surface image inspection apparatus according to the present invention, it is preferable that the feeding speeds of the two flat belt conveyors constituting the V-shaped conveyor are different from each other. (Claim 6)

  According to this configuration, the V-shaped conveyor is a combination of two flat belt conveyors in a V-shape, and one pair of V-shaped conveyors (hereinafter referred to as the left and right for convenience of explanation). Configure the conveyor. The feed speed of the flat belt conveyor may be adjusted independently for each of the left and right sides, or the feed speed may be varied using a link mechanism or a cam mechanism. Here, the feeding directions of the left and right flat belt conveyors may be the same and the speed may be different, or one of them may be moving and the other may be stopped, or one of them may be moving in the opposite direction. . By making a difference in the feeding speed of the flat belt conveyor between the left and right, it is possible to prevent the sorting objects from being conveyed in piles and to separate the sorting objects.

  The spheroid and spherical whole-surface image inspection apparatus according to the present invention is provided with a loading unit that loads the selection target upstream of the transport unit, and includes a dither that applies fine vibration to the loading unit. Yes. (Claim 7)

  According to this configuration, a dither that generates micro vibrations may be provided in the charging unit main body, or a dither that also serves as a chute may be provided at the outlet of the charging unit. The dither works favorably to send the sorting object to the transport section and helps to prevent the sorting object from being piled up. The dither that also serves as a chute acts as a crest prevention plate.

In the spheroid and spherical whole-surface image inspection apparatus according to the present invention, the sorting is performed based on determination results in the input unit, the transport unit, the holding mechanism, the imaging unit, the identification unit, and the identification unit. One or a plurality of basic units having a selection unit for selecting objects is provided. (Claim 8 )

  The number of basic units is determined by the sorting capacity.

In the spheroid and spherical whole-surface image inspection apparatus according to the present invention, the collection unit that collects the selection target that is not separated from the transport unit or the holding mechanism and the selection target that is not separated by the holding mechanism; A recirculation line is provided for feeding the selection objects collected by the collecting means into the charging section. (Claim 9 )

  According to this configuration, sorting objects that are not separated one by one, and sorting objects that are separated from the transport unit or the holding mechanism are collected by the collecting means without human intervention, and the collected sorting objects are returned to the input unit. Since it is refluxed using a line, manpower can be reduced. As the recovery means, a combination of a hopper and a chute can be considered. Moreover, you may comprise a reflux line with a belt conveyor. The sorting object collected by the collecting means may be automatically carried to the input unit via the reflux line.

In the spheroid and spherical whole-surface image inspection apparatus according to the present invention, the width of the upstream V-shaped conveyor in the V-shaped conveyor is not narrower than the width of the downstream V-shaped conveyor. (Claim 10 )

  According to this configuration, if the V-shaped conveyor closest to the charging unit is the first stage for convenience, and the second and third stages are sequentially arranged downstream, the width of the first-stage V-shaped conveyor is preferably the first level. It is narrower than two stages, and the width of the second stage V-shaped conveyor is the same as or narrower than the third stage. Similarly, when there are four or more V-shaped conveyors, the width of the preceding conveyor is wider or the same as that of the next conveyor. As described above, the width of the V-shaped conveyor is determined by the characteristics to be selected and the specifications of the entire surface image inspection apparatus.

In the spheroid and spherical whole-surface image inspection apparatus according to the present invention, the means for binarizing the image and the number of pixels in the contour of the binarized image obtained by the binarization means A size determination unit for determining a size, a unit for obtaining a circumscribed rectangle circumscribing the outline of the selection target from the binarized image, and a shape of the selection target from an aspect ratio of the circumscribed rectangle And shape determining means. (Claim 11 )

  According to this configuration, the contour is a closed region obtained by the binarized image, and the number of pixels included in the contour is set as the size of the selection target. Further, the aspect ratio of the rectangle obtained by approximating the rectangle circumscribing the outline is set as the shape to be selected.

In the spheroid and spherical whole-surface image inspection apparatus according to the present invention, means for calculating a blob from the pixels of the image, and determining the surface condition of the selection target with the number of pixels in the blob as the size of a scratch or the like It is provided with means for determining scratches and the like. (Claim 12 )

  According to this configuration, a blob is a collection of pixels having a certain range of values, and detects a two-dimensional shape that is a lump by isolated point analysis. The blob determination can be performed, for example, by designating a range of gray scale values, and the range can be changed as appropriate. The scratch determination mechanism can identify and determine scratches and dirt present on the surface to be selected.

In the spheroid and spherical whole-surface image inspection apparatus according to the present invention, in the neural network that receives the selection target image, an image determined by the selection standard is used as teacher information, and learning is performed in the neural network. Learning identification means for calculating a degree of sunburn or color unevenness of the selection target using the coupling coefficient. (Claim 13 )

According to this configuration, a selection standard for “tanning” or “color unevenness” is separately determined for the selection target, and the coupling coefficient of the neural network is sequentially corrected using the representative point of this selection standard as teacher information. Ask for. The entire image inspection apparatus of the present invention has an identification mechanism using a neural network by so-called supervised learning. Therefore, the output of the neural network frame has a neural network for sunburn or color unevenness, and outputs identification results for sunburn and color unevenness, respectively. A threshold value may be provided for these identification results, and pass / fail judgment may be performed based on the threshold value.

  In the spheroid and spherical whole-surface image inspection apparatus according to the present invention, the imaging unit includes a combination of an optical filter that realizes a plurality of directivities and a plurality of light sources. (Claim 14)

  According to this configuration, the optical filter that realizes directivity is limited in the direction of light passing through the filter, and helps to reduce the influence of irregularly reflected light entering the camera. The light source may preferably be an LED lamp. Other types of light sources may be used. By appropriately combining and arranging the plurality of light sources and optical filters, it is possible to obtain indirect illumination that is uniform throughout. Indirect lighting plays an important role in obtaining images suitable for computer processing.

  In the spheroid and spherical whole-surface image inspection apparatus according to the present invention, the selection target may be fruits and vegetables such as peppers and lemons. (Claim 15)

  The present invention has the above-described configuration, and by combining a V-shaped belt conveyor and a holding mechanism, it is possible to deal with a plurality of types of sorting objects having various shapes and dimensions. By adopting three or more independently driven V-shaped belt conveyors and a unique holding mechanism, separation of piled-up sorting objects is promoted, and the sorting objects are separated into one by the imaging unit, By imaging the entire surface of each sorting target including the contact portion, it is possible to save labor in sorting work that has been performed manually and to make the sorting work uniform.

It is a systematic diagram which shows the basic composition of the whole surface image inspection apparatus of a spheroid and a spherical surface. It is a top view which shows the external appearance of the conveyance part which consists of a V type belt conveyor. It is a figure which shows the opening angle of a V-shaped belt conveyor. It is a perspective view which shows the outline | summary of the conveyance part comprised with a V-shaped belt conveyor. It is the top view and elevation which show the installation state of a conveyance part. It is an assembly drawing which shows the whole conveyance part containing a holding mechanism. It is a systematic diagram which shows the whole structure of the conveyance part containing a holding mechanism. It is a top view of a holding mechanism. It is explanatory drawing of the holding mechanism provided with the interpolation teg. It is drawing which shows the web camera position in an imaging part. It is drawing which shows the structure and connection of an identification part. It is a flowchart which shows the operation | movement flow of the whole system. It is a flowchart which shows the operation | movement procedure of the conveyance part which is an independent drive system. It is a flowchart which shows the operation | movement procedure of PC for servers of an identification part. It is drawing which shows an image processing procedure. It is a figure which shows the structure of the identification system by a neural network.

  Embodiments according to the present invention will be described below with reference to the drawings.

  For the convenience of explanation of the present invention, red peppers which are fruits and vegetables are taken up as sorting objects. However, the sorting objects are not limited to fruits and vegetables, and generally may be spheroids or spherical bodies, such as industrial products and jewelry such as pearls. The present invention is not limited to the embodiments described below.

(Regarding the configuration of the full-screen image inspection system)
In the description, first, the description will be focused on the device configuration, the operation of each component device will be described after the description of the device configuration, and finally the industrial applicability of the present invention will be mentioned.

  FIG. 1 shows a basic configuration of a spheroid and spherical full-surface image inspection apparatus (hereinafter referred to as a full-surface image inspection apparatus) according to the present invention. The full-surface image inspection apparatus 8 includes a transport unit 2, an imaging unit 4, and an identification unit 5 as its constituent elements. A loading unit 1 for loading the sorting object 9 is provided in the preceding stage of the transport unit 2. The imaging device 4 and the identification unit 5 are connected by the imaging signal line 10, and the image of the selection target 9 captured by the imaging unit 4 is transmitted to the identification unit 5.

  The sorting unit 6 is located at the subsequent stage of the imaging unit 4, and is connected to the identifying unit 5 through the sorting signal line 11, and transmits the determination result in the identifying unit 5 to the sorting unit 6.

  By using an independent drive system, the transport unit 2 separates and transports the selection target 9 up to the imaging unit 4. The conveyance unit 2 is automatically operated using an optical sensor. In the imaging unit 4, in order to inspect the entire surface of the selection target 9, six images are simultaneously captured using six web cameras. As the light source, an LED light source with a diffusion plate provided with a diffusion filter is used in order to suppress specular reflection of the selection target 9. The identification unit 5 identifies the size, shape, and scratches of the selection target 9 by performing image processing.

  Details of the conveyance unit 2, the imaging unit 4, and the identification unit 6 will be sequentially described below.

  The configuration of the transport unit 2 will be described in detail with reference to FIGS. FIG. 2 is a plan view showing an external appearance of the transport unit 2 including the V-shaped belt conveyor 21. As shown in FIG. 3, the V-type belt conveyor 21 is a combination of two flat belt conveyors 20 in a V-shape. The angle θ1 that forms the conveyance surface of the pair of flat belt conveyors 20 is set to, for example, 100 degrees, and the angle θ2 that the conveyance surface of the flat belt conveyor 20 forms with respect to the horizontal plane is inclined to, for example, 40 degrees. For this reason, the sorting object 9 conveyed by the V-shaped belt conveyor 21 is supported by the pair of flat belt conveyors 20 at the bottom of the V groove formed by the flat belt conveyor 20.

  When the selection target 9 is a spheroid such as a red bell pepper, it is possible to align the selection target 9 in the longitudinal direction by combining two flat belt conveyors 20 in a V shape, and imaging that occurs in a later process And image processing is convenient.

  The angle θ1 and the angle θ2 vary depending on characteristics such as the shape of the selection target 9, and are not necessarily limited to 100 degrees and 40 degrees, respectively. For example, in the case of a rectangular industrial product, if the angle θ1 is 90 degrees and θ2 is 45 degrees, the contact area between the selection object 9 and the conveying surface of the flat belt conveyor 20 can be increased, which is convenient for conveying. is there.

  FIG. 4 is a perspective view showing the state of conveyance of the sorting object 9 on the V-shaped belt conveyor.

  As shown in FIG. 2, the transport unit 2 has a configuration in which three V-shaped belt conveyors 21 described above are connected in series. That is, the end portion 25b of the first-stage V-shaped conveyor 25 close to the charging section 1 is disposed at the start end portion 26a of the second-stage V-shaped conveyor 26 with a step, and the end portion 26b of the second-stage V-shaped conveyor 26 is It is arranged at the start end 27a of the third stage V-shaped conveyor 27 with a step. The end portion 27 b of the third stage V-shaped conveyor 27 is connected to the holding mechanism 3 that forms a part of the transport unit 2. Details of the holding mechanism 3 will be described later.

  FIG. 5 is a system diagram showing the assembled state of the V-shaped belt conveyor 21. The upper diagram of FIG. 5 is a plan view and the lower diagram is an elevation view.

  The V-type belt conveyor 20 is installed with an inclination as shown in the lower diagram of FIG. 5, and the V-type conveyors 25, 26, and 27 at each stage are independently geared motors by conveyor drive mechanisms 29a, 29b, and 29c. (Not shown). That is, the conveyor drive mechanisms 29a, 29b, and 29c have independent and independent drive mechanisms, and are individually driven and controlled by a control device (not shown).

  The sorting object 9 is conveyed while moving upward from the start end portion 25a of the first-stage V-shaped conveyor 25 toward the end portion 27b of the third-stage V-shaped conveyor 27. When the sorting object 9 is fruits and vegetables, the V-shaped conveyors 25, 26, and 27 can be adjusted within a range of inclination angles of 20 degrees to 30 degrees. When the selection target 9 is red pepper, it is preferably installed at 30 degrees.

  Since the V-shaped conveyors 25, 26, and 27 are installed with an inclination in the upward direction as described above, when a large amount of the selection target 9 is input from the input unit 2 to the transport unit 2, the selection target 9 that is in a lump shape. Collapses in the conveying process, and the separation of the sorting object 9 is promoted.

  Therefore, the inclination angles of the V-shaped conveyors 25, 26, and 27 are appropriately determined according to the shape, surface characteristics, and the like of the selection target 9. Degrees to 45 degrees are an appropriate range.

  As shown in FIGS. 5 and 6, optical sensors 32, 33, and 34 are provided in the V-shaped conveyors 25, 26, and 27 in the vicinity of the terminal portions 25b, 26b, and 27b in a direction perpendicular to the conveying direction. The signals from the optical sensors 32, 33, and 34 are connected to a control device (not shown), and control for stopping the operation of the V-shaped conveyors 25, 26, and 27 is performed. The optical sensors 32, 33, and 34 are transmissive sensors using LEDs as issuing elements, and detect the presence or absence of an object between them in cooperation with the light receiving element on the opposite side of the light emitting element. The optical sensor may be a reflective type. In the reflection type sensor, a reflector is placed on the opposite side of the light emitting element, and the presence or absence of an object is detected by a light receiving element arranged in the same unit as the light emitting element. Other types of proximity switches (for example, those using limit switches) may be used instead of the optical sensors.

  The width of the second-stage V-shaped conveyor 26 is narrower than the width of the first-stage V-shaped conveyor 25, and the width of the third-stage V-shaped conveyor 27 is narrower than that of the second-stage V-shaped conveyor 26. . This is advantageous for breaking the piled up state as the sorting target 9 that has been put into the first stage V-shaped conveyor 25 is transported.

  In this embodiment, the reduction ratio of the width of the V-shaped conveyor is 30%. The optimum value varies depending on the shape, characteristics, etc. of the selection object 9 as to how narrow it is, and the width of the third stage V-shaped conveyor 27 is 70% or more of the width of the second stage V-shaped conveyor 26. May be optimal, and it may be preferable that there is little difference between the widths of the second-stage V-shaped conveyor 26 and the third-stage V-shaped conveyor 27.

  The input unit 1 may include a dither device 13 (see FIG. 5). The dither device 13 may be composed of, for example, an electrostrictive vibrator or a piezoelectric element, and may generate micro vibrations by applying a voltage. The sorting object 9 is prevented from becoming a lump and acts as a feeder for sending the sorting object 9 to the transport unit 2.

  Instead of providing the dither device 13 in the main body of the input unit 1, a pile prevention plate 14 may be provided at the outlet of the input unit 1 or the entrance of the transport unit 2 (see FIG. 6). The pile prevention plate 14 may be one that generates micro-vibration by applying a voltage with an electrostrictive vibrator or a piezoelectric element. FIG. 6 is a system diagram showing an assembled state of the transport unit 2 including the input unit 1 and the holding mechanism 3, and the upper diagram of FIG. 6 is a plan view and the lower diagram is an elevation view.

  A holding mechanism 3 is provided upstream of the three-stage V-shaped conveyors 25, 26 and 27. That is, as shown in FIG. 6, the holding mechanism 3 is arranged with a step at the end portion 27 b of the third-stage V-shaped conveyor 27. The holding mechanism 3 mainly includes a pulley 43, a pulley driving device 44, and a teg 41, and the selection target 9 that has been transported through the third-stage V-shaped conveyor 27 is transferred to the holding mechanism 3. (See FIG. 7). That is, the holding mechanism 3 is composed of two tegs 41 that are stretched in parallel and horizontally, and a set of pulleys 43 that suspend and drive the tegs 41 to rotate. Since the distance between the two tegs 41 is smaller than the dimension in the short direction of the sorting object 9 and is stretched with an appropriate distance, the sorting object 9 is stably positioned on the tegus 41. In the present embodiment, the distance between the tegs 41 is set to 25 mm from the representative dimension of the red bell pepper that is the selection target 9, but is not limited thereto.

  As shown in FIG. 6, each Tegg 41 is suspended and driven by three pulleys 43, and each pulley 43 is driven synchronously by a pulley driving device 44, so that the two Tegs 41 are the same. Driven at a speed, the sorting object 9 is conveyed in a direction away from the third stage V-shaped conveyor 27. The pulley drive mechanism 44 has a separate drive mechanism independent from the conveyor drive mechanism 29, and is individually driven and controlled by a control device (not shown). A control device (not shown) may be preferably common to the pulley drive mechanism 44 and each conveyor drive mechanism 29. If the control device is shared, the equipment can be simplified. FIG. 8 shows a plan view of the holding mechanism 3. The sorting object 9 is conveyed from the bottom to the top in FIG. In the present embodiment, the tegs 41 used as a fishing line is adopted as the holding thread, but it is sufficient if it does not hinder the entire surface imaging when imaging the selection target 9, and a light-transmitting thread is preferable. . Even if it does not have optical transparency, the diameter may be sufficiently small to have a strength capable of transporting the selection target 9 and to be sticky to the pulley 43. This is because there is a case where there is no problem in image processing depending on the selection target 9 as long as the thread is thin even if it is not light transmissive.

  In the holding mechanism 3, since the selection target 9 is held on the two tegs 41, the structure is suitable for imaging the bottom surface of the selection target 9 from below. That is, since the selection target 9 is supported by two light-transmitting yarns, there is nothing that prevents imaging.

  As shown in FIG. 9, an interpolation tag 45 may be provided between the two tags 41. FIG. 9 is a plan view (a) and an elevation view (b) showing the positional relationship between the selection target 9 in the holding mechanism 3 in the case where the interpolated tex 45 is provided, the tegus 41, and the auxiliary gusset 45. If the interpolated tess 45 are provided, the selection target 9 can be supported at three points in cooperation with the two tegs 41 on both sides. If the three points are supported, the posture of the selection target 9 is stable and imaging is easy, and imaging can be performed smoothly. In addition, when the selection target 9 is a spheroid as in the present embodiment, the longitudinal direction is aligned with the transport direction, and image processing that occurs in a subsequent process becomes easy. Further, as shown in FIG. 9A, even when the selection target 9b is slightly smaller than a separately defined standard, it can be transported by the holding mechanism 3, and the entire image inspection apparatus 8 performs the selection. It can be carried out.

  The pulley 43 that drives the two Tegs 41 is driven coaxially by the pulley drive device 44 via the pulley drive shaft 42, and the remaining pulley 43 that is not driven by the pulley drive device 44 suspends the Tegs 41. On the other hand, the small pulley 47 is not coupled to the pulley drive shaft 42, and idles according to the movement of the interpolating tex 45.

  The holding mechanism 3 is installed in the imaging unit 4 (see FIG. 1). The imaging unit 4 is provided with six Web cameras 51, eight LED light sources 53, and eight diffusion filters 55. The pair of LED light sources 53 and the diffusion filter 55 cooperate to form one indirect light source. The light from the LED light source 53 is dispersed by the diffusion filter 55 and gives indirect illumination to the selection target 9. The diffusion filter 55 only needs to diffuse light. For example, the diffusion filter 55 may be a plate member having a sawtooth cross section made of glass or synthetic resin. The indirect light source prevents specular reflection on the surface of the selection target 9 and is therefore suitable for identifying and determining wrinkles on the surface such as scratches on the selection target 9.

  In the present embodiment, four of the six web cameras 51 are shifted by 45 ° from the horizontal plane around the side of the selection target 9 as viewed from the transport direction every 90 ° as shown in FIG. As shown in FIG. 10 (b), the remaining two cameras are arranged with a slight angle with respect to the transport direction before and after the selection target 9, and the six Web cameras 51 constitute an orthogonal coordinate system. It is arranged to do. Then, the entire web of the selection target 9 held on the holding mechanism 3 is imaged by the six Web cameras 51.

  The Web camera 51 includes UCAM-E1L30MNWH manufactured by ELECOM (maximum resolution: 640 × 480 pixels, image receiving element: 1/4 inch, 300,000 pixel CCD sensor, quantization: 24 bits / pixel, resolution: 0.3 mm / pixel). In addition, as the indirect light source, eight PM-L200 (W) made by LED light source ELPA with diffusion plate were used.

  Next, the configuration of the identification unit 6 will be described with reference to FIG. The identification unit 6 includes one server personal computer 92 (hereinafter referred to as a PC) and two client PCs 91.

  In the identification unit 6, six Web cameras 51 are connected to the server PC 92 and the client PC 91 via the USB interface. That is, two Web cameras 51 are connected to each PC, and the images captured by the imaging unit 4 are transferred from the client PC 91 to the server PC 92 via the communication line 93, and the six images captured are captured. Using the image, the server PC 92 identifies the size, shape, and scratches of the selection target 9, and identifies sunburn and color unevenness.

  In the entire surface image inspection apparatus 8 according to the present invention, the holding mechanism 3 is provided with a collecting means (not shown) for collecting the selection target 9 that is separated from the transport unit 2 or the holding mechanism 3 or the selection target 9 that is not separated by the holding mechanism 3. A reflux line (not shown) may be provided at a subsequent stage for transporting the selection target 9 collected by the collecting means (not shown) to the input unit 1.

  If the reflux line is provided in this way, the sorting objects 9 that are not separated one by one and the sorting objects 9 that are separated from the transport unit 2 and the holding mechanism 3 are collected without human intervention, and the collected sorting objects are collected. Since 9 is returned to the charging unit 1, manpower can be reduced. As the recovery means, a combination of a hopper and a chute can be considered. Moreover, you may comprise a reflux line with a belt conveyor. The sorting object 9 collected by the collecting means may be automatically carried to the input unit 1 via the reflux line.

(About overall operation of full-screen image inspection system)
The basic configuration of the full-surface image inspection apparatus 8 according to the present invention, the transport unit 2, the imaging unit 4, and the identification unit 5 have been described above. Next, these operations will be described. First, the overall flow of the entire image inspection apparatus 8 according to the present invention will be described with reference to the flowchart of FIG.

  1) A plurality of sorting objects 9 are input from the input unit 1 to the transport unit 2.

  2) 3) The inputted sorting object 9 is separated into one by the three V-shaped conveyors 25, 26, 27 and the tegs 41 of the holding mechanism 3 that are independently driven, and is carried to the imaging unit 4. The operation (2) 3)) of the transport unit 2 will be described in detail separately.

  4) When the sorting object 9 comes to a substantially intermediate position of the holding mechanism 3 (in the imaging unit 4) by the feed of the Tegs 41, the sorting object 9 stops, and the sorting object 9 held by the holding mechanism 3 becomes 6 Images are picked up simultaneously from six directions by a single Web camera 51.

  5) The six captured images are sent to the server PC 92 and the client PC 91 of the identification unit 5, and when the six images are aligned, the size of the selection target 9 is determined by the server PC 92 of the identification unit 5. Identify the shape and scratches.

  6) Further, the six images are identified for “sunburn” and “color unevenness” by template matching using a neural network in the server PC 92 of the identification unit 5. The operation (5) 6)) of the identification unit 5 will be described in detail separately.

  7) Based on the determination result in the identification unit 5, the selection unit 6 downstream of the identification unit 5 performs selection according to the pass / fail determination. The selection object selected by the selection unit 6 in response to the pass determination flows to the shipping line. On the other hand, the rejected selection object is conveyed from the selection unit 6 to the input unit 1 via a reflux line (not shown).

(Details of operation of transport unit 2)
The operation procedure of the transport unit 2 including the holding mechanism 3 will be described with reference to FIG. These operation procedures are computer-controlled by a control device (not shown).

First, all the V-shaped conveyors 25, 26, and 27 and the tegs 41 are rotated (forward rotation) in the transport direction. Next, when the optical sensor 1 (32) reacts, it recognizes that the selection object 9 has been conveyed to the second stage V-shaped conveyor 26, and stops the first stage V-shaped conveyor 25. However, the optical sensor 1 (32) if continued to react _{for 1} second t, forward rotation of the first stage V-type conveyor 25 again t ₂ seconds.

Next, when the optical sensor 2 (33) reacts, it is recognized that the selection target 9 has been conveyed to the third stage V-shaped conveyor 27, and the second stage V-shaped conveyor 26 is stopped. However, the optical sensor 2 (33) if continued to react _{for 1} second t, forward rotation of the second stage V-type conveyor 26 again t ₂ seconds. Here, the values of t ₁ , t ₂ , and x can be appropriately changed and adjusted by the selection target 9.

  Next, when the optical sensor 3 (34) reacts, it recognizes that the selection object 9 has been transported to the tegs 41, and stops the third stage V-shaped conveyor 27. Next, after the Tegg 41 rotates positively for 1 second, it is stopped for 1 second at an intermediate position of the Tegg 41 for imaging. Finally, the Teggs 41 is rotated forward again to discharge the selection target 9. If all the optical sensors 32, 33, and 34 do not react for 1 second, the conveyance is automatically stopped. It is not limited to these values, and can be changed and adjusted as appropriate depending on the selection target 9.

  Next, the operation of the transport unit 2 will be described in relation to the selection target 9.

  Since the normal inspection cannot be performed in the state where the selection targets 9 are adjacent to each other, the selection target 9 needs to be separated into one by the imaging unit 4. Therefore, the transport unit 2 uses an independent drive system that drives the three V-shaped belt conveyors 20 and the tegs 41 independently and in conjunction with each other. The independent drive system will be specifically described below.

  The inputted sorting object 9 is piled up, but is carried to the imaging unit 4 by the three V-shaped conveyors 25, 26, 27 and the tangs 41 of the holding mechanism 3. The two flat belt conveyors 20 constituting the V-shaped belt conveyor 21 are controlled separately. In other words, there are a “forward feed mode” in which two flat belt conveyors 20 are driven at the same speed in the same direction and a “mountain break mode” in which the flat belt conveyor 20 is driven at different speeds. There are a mode in which the direction is the same and the speed is changed, a mode in which the feeding direction is different, a mode in which one flat belt conveyor is stopped, and the like. By a program control in a control device (not shown), the “forward feed mode” and the “hill-climbing mode” are repeated as appropriate, and the selection target 9 is conveyed upward.

  Furthermore, since the V-shaped belt conveyor 21 is installed with an upward inclination, the sorting object 9 stacked on the sorting object 9 collapses in the counter-conveying direction at the time of transportation, which is advantageous for the mountain collapse (separation) of the sorting object 9. To work.

  Further, since the V-shaped conveyors 25, 26, 27 and the holding mechanism 3 are respectively connected with a step, the selection target 9 is separated by a drop when moving to the next step.

  Furthermore, since the V-shaped conveyors 25, 26, and 27 and the holding mechanism 3 are driven independently, their feed rates can be adjusted separately. It is preferable that the feed speed of the second stage V-shaped conveyor 26 is higher than the feed speed of the first stage V-shaped conveyor 25. By increasing the feed speed, the lump of the selection target 9 that has dropped onto the second-stage V-shaped conveyor 26 collapses. Similarly, it is preferable that the feed speed of the third stage V-shaped conveyor 27 is higher than the feed speed of the second stage V-shaped conveyor 26. Thereby, separation of the selection object 9 is promoted. Furthermore, it is preferable to make the feed speed of the tees 41 of the holding mechanism 3 faster than the feed speed of the third-stage V-shaped conveyor 27, and the same effect can be expected, and the separation of the sorting object 9 is promoted.

  Since the conveyor is V-shaped, the sorting object 9 is conveyed to the bottom of the V-shaped belt conveyor 21 with its direction aligned in the longitudinal direction.

  Since the selection target 9 is separated and imaged in the longitudinal direction, the image processing in the identification unit is simplified and the identification capability is enhanced. This is because it is often difficult to identify an object that is not separated.

(Details of operation of imaging unit 4)
The Web camera 51 constantly monitors the imaging unit 5 and determines that the selection target 9 has arrived at the imaging position by taking a difference from an image when the selection target 9 does not exist in the holding mechanism 3 and holds it. The feeding of the mechanism 3 is stopped, and six surfaces of the selection target 9 are simultaneously imaged by the six web cameras 51 in a stationary state. The image captured by the client PC 91 is transferred to the server PC 92 via the LAN.

  The determination of the imaging position is performed on the monitor image captured once every 2 seconds, and when the selection target 9 is continuously detected, it is determined that the selection target 9 has arrived at the imaging position. Imaging may be performed while the selection target 9 is moving without stopping the holding mechanism 3. If imaging is performed while the selection target 9 is moving, the operating rate of the full-screen image inspection apparatus is improved.

  When obtaining the difference between the monitor images, it is possible to determine whether the identification target is a red pepper or a blue pepper by using the color information of the image. Using the color information of the red bell pepper, when the hue of the monitor image continuously matches the color information of the red bell pepper, it may be recognized that the selection target 9 has been transported to the imaging position. In this case, not only the imaging position but also the type of the selection target 9 is determined.

  When the type of the selection target 9 is specified in advance, a predetermined time may be measured with a timer after the optical sensor 3 (34) detects the selection target 9, and it may be determined as the imaging position. Note that the imaging position is planned to be approximately the middle position of the holding mechanism 3.

(Details of operation of identification unit 5)
The inspection procedure (operation) in the identification unit is shown below.

  In the present embodiment, the appearance quality inspection is performed by performing the following processing for each of the six captured images of the six red peppers. Here, a flowchart of the inspection process for each image of the selection target 9 executed by the server PC 92 in the identification unit 4 is shown in FIG. 14, and a flow of the image process is shown in FIG.

  1) Wait until six images (six images) are aligned with the images captured by the client PC 91.

  2) Convert the captured RGB color image into a CIE 1976 (L *, a *, b *) color image.

  3) Binarization is performed so that only a region (pepper region) where the Euclidean distance from the template color of red bell pepper (here, red) is equal to or less than a predetermined threshold value (“green pepper region”) is converted into a monochrome image.

  4) Eight neighborhood contraction / expansion processing is performed on each of the left, right, up, down, and diagonal neighborhoods to remove noise.

  5) The area of each connected area consisting of the pixel value “1” is calculated by the labeling process, and the pepper area is extracted by extracting the connected area whose area is equal to or greater than a predetermined threshold. After that, each determination process for the size, shape, and black scratch described below is performed.

  6) The size determination is performed by determining “large” when the area of the bell pepper area calculated in 5) is equal to or larger than a threshold determined based on the standard, and determining “small” otherwise. .

  7) In the shape determination, first, a rectangle having the smallest area among rectangles in contact with the pepper region is detected to perform a rectangle approximation.

  8) Next, in the approximated rectangle, the shape is determined as “non-defective” only when the long side / short side is equal to or greater than the threshold value determined based on the standard, and otherwise determined as “defective”. Make a decision.

  9) Scratch determination is performed by first converting a captured RGB color image into a grayscale image including only an R value, and performing binarization by setting only an area (black scratch area) equal to or less than a predetermined threshold value to “1”. AND operation is performed on the monochrome image obtained by the above and the monochrome image having the pixel value “1” only in the bell pepper area extracted in the above 5) to obtain a scratch image.

  10) Next, labeling processing is performed on the scratch image to calculate the area of the connected region having the pixel value “1”, and it is determined as “non-defective” only when the area is equal to or less than the threshold value determined based on the standard. By determining “scratch defect”, scratch determination is performed.

  By integrating the determination results of all six images as follows with the determination results of each image obtained as described above, the final inspection result is obtained.

  Regarding size determination, “small” is set when there is an image determined to be “small”, and “large” is set otherwise. Regarding shape determination, a case where there is an image determined to be “shape defect” is defined as “shape defect”. In addition, regarding scratch determination, a case where there is an image determined as “scratch defect” is defined as “scratch defect”. Finally, it is determined as “non-defective” only when it is neither “shape defect” nor “scratch defect”.

  Next, the identification of “sunburn” and “color unevenness” by template matching using a neural network (hereinafter referred to as NN) will be described. The NN is suitable for identifying ambiguous defect situations such as actual sunburn and ripeness determined by human eyes.

  The captured red bell pepper images are sent to the server PC 92, wait for the six images to be aligned with the server PC 92 of the identification unit 5, are binarized by a predetermined threshold, and have a white background. It is converted into an image (the same processing as 1), 2) and 3) in FIG. 14 is performed). Then, in order to convert the binarized image data into numerical values suitable for the input of the neural network, a frequency analysis process is performed using a two-dimensional fast Fourier transform (hereinafter also referred to as 2DFFT), and an addition average is obtained. The slab value is input to each unit (also referred to as 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 the present embodiment, the NN has a three-layer structure, and the three-layer structure calculation unit includes three layers: an input layer (64 neuro elements), a hidden layer (35 neuro elements), and an output layer (1 neuro element). It is made up of. In the input layer, the slab value calculated for each effective area of the component obtained from 2DFFT 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. Each output layer is provided with one neuro element so as to correspond one-to-one for each type of “red pepper quality” to be identified. Then, the response values of the output neuro elements based on the weighting factor between the neuro elements completed by learning are calculated for several output neuro elements (here, one).

  Here, the basic configuration of the NN in the identification unit 5 is shown in FIG. Each neuro template in FIG. 16 is “sunburn” and “color unevenness”, which are specific qualities of red peppers, and these are individually identified. In the identification process, the red pepper image is subjected to 2DFFT processing, which is preprocessing, and a slab value that 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.

  When a (correct) tanning image conforming to the grade standard is input, the response value of the unit of the output layer of the tanning neuro template is close to the maximum value. The scale of the reaction value may be arbitrary, but here, for convenience of determination, a sigmoid function is applied to the calculation result of the output layer, and the result is normalized to 0 to 1 and output.

  Similarly, for the output layer of the color unevenness neuro template, a value close to 1 is output when a correct tan image is input, and a value less than 1 is output for an incorrect tan image depending on the degree. .

  With this template matching by NN, highly accurate identification determination can be realized.

  By the way, in the whole image inspection apparatus for red bell peppers according to the embodiment of the present invention, quality data based on image data previously captured by a digital camera for red bell peppers to be selected 9 is learned and registered using NN. Yes. Of the image data, data relating to a characteristic pattern (pattern) that is particularly important for identifying the quality of red bell pepper is subjected to frequency analysis processing using a two-dimensional fast Fourier transform, and an input pattern (slab value) of NN Input and stored in the input layer. Then, the image data captured by the Web camera 51 of the image capturing unit 4 is sequentially converted by the above-described procedure, and an input pattern to the NN is created. The created input pattern is transmitted and processed from the input layer → the hidden layer → the output layer, and a reaction value as an output is output from the output layer. This output value is based on the weighting coefficient obtained by learning so far, and in the case of this example, the degree of conformity with the standard of red pepper quality is output. For this reaction value, a pass / fail determination is made with a threshold value determined based on the standard, and a final discrimination determination result is obtained.

On the other hand, when the correct output value is given to the correct image in accordance with the standard for the quality pattern of red peppers in the NN, it is transmitted and processed in the order of output layer → hidden layer → input layer, Learning is performed on the weighting factor. The learning of this weighting factor is to reduce the difference between the actual output value and the correct output value, and to reduce the coupling between the neuron elements of the input neuron element, hidden neuron element, and output neuron element. This is done by changing and converging the strength.
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, η is a learning constant, α is an inertia constant, and β is a vibration constant.

  According to the entire image inspection apparatus of the present invention, it is possible to automate the inspection of fruits and vegetables by inspecting the entire image of the spheroid and sphere. It can also be used for full-surface image inspection of industrial products and jewelry.

DESCRIPTION OF SYMBOLS 1 Input part 2 Conveying part 3 Holding mechanism 4 Imaging part 5 Identification part 6 Sorting part 8 Whole surface image inspection apparatus (basic unit)
DESCRIPTION OF SYMBOLS 9 Selection object 10 Imaging signal line 11 Selection signal line 12 Input port 13 Dither device 14 Stack prevention board 15 Collecting means 16 Reflux line 20 Flat belt conveyor 21 V type belt conveyor 25 1st stage V type conveyor 26 2nd stage V type conveyor 27 Third-stage V-shaped conveyor 29 Conveyor drive mechanism 32 Optical sensor 1
33 Optical sensor 2
34 Optical sensor 3
41 Tegs 42 Pulley drive shaft 43 Pulley 44 Pulley drive mechanism 45 Interpolation Tegs 47 Small pulley 51 Web camera 53 LED light source 55 Diffusion filter 91 Client PC
92 PC for server
93 Communication line 94 USB interface

Claims (15)

A transport unit for transporting a selection object of a spheroid or a spherical body;
A holding mechanism for holding the sorting object following the transport unit;
An imaging unit having a plurality of cameras for imaging the selection target;
A spheroid and spherical full-surface image inspection apparatus including an identification unit that identifies and discriminates a plurality of images to be selected captured by the imaging unit;
The transport unit has a plurality of V-shaped conveyors in which two flat belt conveyors are combined so that the cross section in the belt traveling direction is V-shaped,
Each of the plurality of V-shaped conveyors has an independent drive mechanism, and is arranged with the steps lowered in order from the upstream to the downstream in the conveying direction, and each of the V-shaped conveyors conveys the sorting object upward. It is installed with an inclination like
The holding mechanism includes a plurality of string-like holding units for conveying and holding the selection target, and is driven independently of the V-shaped conveyor, and a starting end portion of the string-like holding unit is an outermost part of the conveying unit. An apparatus for inspecting the entire surface of a spheroid and a spherical body arranged at a terminal portion of a lower V-shaped conveyor.   The entire image inspection apparatus for spheroids and spherical bodies according to claim 1, wherein the string-like holding part has light permeability.   3. The spheroid and spherical full-surface image inspection apparatus according to claim 1, wherein the number of cameras is six, and the plurality of images are picked up from six different directions of the selection target. 4.   An imaging position detection unit that detects that the selection target has reached a predetermined position of the holding mechanism, and a synchronous imaging mechanism that receives an instruction from the imaging position detection unit and simultaneously images the selection target by the camera. The whole surface image inspection apparatus of the spheroid and spherical body of any one of Claims 1-3. At least an optical sensor is attached to the starting end of each V-shaped conveyor excluding the uppermost stage and the terminal end of the lowermost V-shaped conveyor,
When it is detected by the optical sensor that the object to be sorted has been conveyed from the previous stage to the next V-shaped conveyor or from the lowermost V-shaped conveyor to the holding mechanism, the previous or lowermost V-shaped conveyor is detected. Means for stopping the conveyance of
Alternatively, when it is detected by the optical sensor that the object to be sorted has been conveyed from the previous stage to the next V-shaped conveyor or from the lowermost V-shaped conveyor to the holding mechanism, the V or lowermost stage V is detected. Means to reverse the conveyance of the mold conveyor for a certain time,
The whole surface inspection apparatus of the selection object of any one of Claims 1-4 provided with.   The spheroid and spherical surface image inspection apparatus according to any one of claims 1 to 5, wherein the feeding speeds of the two flat belt conveyors constituting the V-shaped conveyor are different from each other.   The spheroid according to any one of claims 1 to 6, further comprising a dither that feeds the sorting object upstream of the transport unit, and further includes a dither that imparts a slight vibration to the throwing unit. Full-surface image inspection device for spherical bodies. One or a plurality of basic units including the input unit, the transport unit, the holding mechanism, the image capturing unit, the identification unit, and a selection unit that selects the selection target based on the determination result in the identification unit. The entire surface image inspection apparatus for a spheroid and a spherical body according to claim 7 . A collecting means for collecting a sorting object that is separated from the transport unit or the holding mechanism and a sorting object that is not separated by the holding mechanism, and a recirculation flow for feeding the sorting object collected by the collecting means to the input unit 9. The entire image inspection apparatus for spheroids and spherical bodies according to claim 7 or 8 having lines. Entire image inspection device of the spheroid and the spherical body according to any one of the V-width of the V-type conveyor upstream stage is not smaller than the width of the V-type conveyor downstream stage in a conveyor according to claim 1-9. Means for binarizing the image;
A size determination unit that determines the size of the selection target from the number of pixels in the contour of the binarized image obtained by the binarization unit;
Means for obtaining a circumscribed rectangle circumscribing the contour of the selection target from the binarized image;
Entire image inspection device of the spheroid and the spherical body according to any one of claim 1 10, from the aspect ratio of the bounding rectangle and a determining shape determination means the shape of the screened object. Means for calculating a blob from pixels of the image;
The spheroid and spherical body according to any one of claims 1 to 11 , further comprising a flaw determination unit that determines the surface state of the selection target by using the number of pixels in the blob as a flaw size. Full-screen image inspection device. In the neural network having the selection target image as an input, the image determined by the selection standard is used as teacher information, the coupling coefficient in the neural network is obtained by learning, and the selection target tanning is performed using the coupling coefficient. Alternatively, the entire image inspection apparatus for a spheroid and a spherical body according to any one of claims 1 to 12 , further comprising a learning identification unit that calculates a degree of color unevenness.   The spheroid and spherical body full-surface image inspection device according to claim 1, wherein the imaging unit includes a combination of an optical filter that realizes a plurality of directivities and a plurality of light sources.   The spheroid and spherical body entire surface image device according to any one of claims 1 to 14, wherein the selection target is fruits and vegetables such as bell pepper and lemon.