JPH04364449A - Fruit defect detecting apparatus - Google Patents

Abstract

PURPOSE:To obtain a fruit defect detecting apparatus for detecting the defective part of a material to be processed having the different colors from the surrounding parts such as flaws, sunburn and rust for the material to be processed such as an apple. CONSTITUTION:The image information which is picked up with an image sensing means 2 is divided into the specified regions comprising the specified number of the images. A predetermined evaluating value is computed for every region, and it is judged whether a detect in present in a material to be processed 6 based on the computer value with a processing means 4. The processing means 4 has the parts for performing the following functions. An image processing means 45 computers the evaluating value based on the information from image memories 42. A neural network 44 judges to which defect the region belong to based on the evaluating-value data from the image processing means 45. A first judging means for the detect forms the distribution map of each defect described above based on the result of the judgment of the judging neutral network 44 and performs the final judgment by using the distribution map.

Description

[Detailed description of the invention]

0001

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a fruit defect detection apparatus for determining the hue of a target object, such as an apple, which has a complex hue.

0002

2. Description of the Related Art Conventionally, a fruit defect detection apparatus includes an imaging means for imaging an object to be processed, an image taken by the imaging means, which is divided into pixels, and a red component, a green component, It has a hue determination processing means that evaluates the hue of each pixel by taking the absolute values of the hue elements of the blue component and their ratios, and performs a comprehensive evaluation of the overall hue from this evaluation. (Refer to JP-A-63-200279)

[0003] Other fruit defect detection devices include:
There is a method that uses an infrared camera to detect the temperature difference between an undamaged part and a scratched part of the object to be processed.

0004

[Problem to be Solved by the Invention] However, the problem with fruit defect detection devices that perform comprehensive evaluation of hue is that they evaluate a certain average color on the surface of the pericarp, and cannot extract specific colors. There is.

[0005] Furthermore, a device for detecting flaws on an object to be processed using an infrared camera has a problem in that the device is expensive.

SUMMARY OF THE INVENTION Therefore, an object of the present invention is to provide a fruit defect detection device for detecting defective parts of an object to be processed, such as apples, that have a different color from the surrounding area, such as scratches, sunburn, and rust. It's about doing.

[0007]

[Means for Solving the Problems] According to the present invention, there is provided an imaging means for outputting image signals obtained by imaging objects to be processed having different hues, and image information taken by the imaging means, which is composed of a predetermined number of images. processing means for dividing the object into regions, calculating a predetermined evaluation value for each region, and determining the presence or absence of a defect in the object to be processed based on the calculated value; An image memory for latching and storing, an image processing means for calculating the evaluation value based on information from the image memory, and a neural system for determining which defect the area belongs to from the evaluation value information from the image processing means. network, and means for creating a distribution map of each defect for each object to be processed based on the determination result of the neural network, and final determination of defects using the distribution map, A fruit defect detection device is provided, characterized in that when determining a target to be processed, characteristics of the defective portion are collected as learning data using a plurality of types of target to be processed for learning. It will be done.

[0008]

[Operation] In the operation during fruit defect learning, a fruit with a defective portion is placed on the conveying means and the fruit arrives within the imaging range of the imaging means. When the detector detects the conveying means, the processing device inputs and latches the images from the imaging means into the image memory separately as a red image, a green image, or a blue image. An image of the defective part is manually extracted from the captured image. Divide the cut out image into sub-regions made up of pixels, calculate the evaluation value for each sub-region,
Use as defect learning data. The operation is repeated until a plurality of pieces of learning data are collected for each defect. Next, the learning data is used in a neural network circuit to learn the characteristics of each defect. After completing this operation, next perform the operation for fruit determination. First, fruits are placed on the conveying means. The conveying means moves and arrives within the imaging range of the imaging means. At this time, the detector is arranged to detect the conveyance means. When the detector detects the conveying means, the processing device inputs the images from the imaging means into the image memory separately as a red image, a green image, or a blue image and latches them. Divide the captured image into sub-regions made up of pixels,
An evaluation value is calculated for each sub-region, and then a neural network circuit determines the attributes of each sub-region. A map of each defect is created based on the judgment results of the neural network circuit. The size and number of defect maps are determined from the defect map, and a final judgment is made.

[0009]

DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 is a block diagram showing an embodiment of a fruit defect detection apparatus according to the present invention. FIG. 2 shows a specific example of a processing device used in the fruit defect detection device. Note that this embodiment will be explained using a fruit such as an apple as the object to be processed.

Referring to each figure, the fruit defect detection device is as follows:
It has a conveyance means for conveying the fruit, and an imaging means for taking an image of the fruit 6 conveyed by the conveyance means. The conveyance means has a pallet 1 which can be moved on a fruit sorting line on which fruits 6 are placed and sorted. The imaging means has a camera 2 that takes an image of the fruit 6 on the pallet 1. The camera 2 is a color CCD or the like, which divides a captured image into a red component, a green component, or a blue component for each pixel, and transmits the divided image as a red image, green image, or blue image to a processing device to be described later. In addition, the fruit defect detection device
It has a detector 3 that detects when the fruit 6 comes within the range of the camera 2, and a processing device 4 that detects a defective part on the surface of the fruit 6 based on the image information sent from the camera 2. ing.

Furthermore, the fruit defect detection apparatus includes an illumination device 5 for illuminating the fruit in order to keep the illumination conditions for camera imaging constant, and a terminal device 7 for inputting and outputting processing data.

The processing device 4 includes a plurality of A/D converters 41 for converting image information such as a red image, a green image, or a blue image sent from the camera 2 into digital information, and a plurality of A/D converters 41 for latching and storing image information. It has an image memory 42 and a gate circuit 43 that outputs imaging timing based on the output signal from the detector 3. Furthermore, the processing device 4 includes a neural network 44 to which information in the image memory 42 is transferred, and an image processing device 45 that makes a final determination based on the results of the neural network 44. For example, a microcomputer is used as the image processing device 45. The neural network 44 determines the presence or absence of defects in the fruit 6 based on information from the image processing device 45. Further, the neural network 44 may be a microcomputer as long as it can execute a neural network algorithm.

Next, the operation of determining defects in fruit 6 of a new variety using the fruit defect detection device will be described below. When determining defects in fruit 6 for the first time, perform the operations described in "Operations when learning fruit defects" below,
Let the neural network learn the characteristics of each defective part.

``Operations during fruit defect learning'' (1) Place the fruit 6 with the defect on the pallet 1 on the fruit sorting line. (2) The pallet 1 carrying the fruit moves and arrives within the imaging range of the camera 2. At this time, the detector 3
is arranged to detect. (3) When the detector 3 detects the pallet 1, the processing device 4
The images captured by the camera 2 are input into the image memory 42 separately as a red image, a green image, or a blue image, and are latched and recorded. (4) From the captured image, the image of the defective part is manually cut out into small cutout areas as shown in FIG. (5) Cut out the image by k×1 (1≦k≦N/
2, 1≦k≦M/2: N and M are the number of vertical and horizontal images in the image memory, respectively. It is calculated and used as defect learning data. Examples of evaluation values include RGB average values and statistics in other color spaces. (6) Repeat operations (1) to (5) until about 5 to 10 pieces of learning data are collected for each defect. (7) Using the learning data, make the neural network circuit learn the characteristics of each defect.

[0015] After the above-mentioned ``operations during fruit defect learning'' are completed, the following ``operations during fruit determination'' are performed. "Operations when judging fruit" (1) Place the fruit on pallet 1 on the fruit sorting line. (2) The pallet 1 carrying the fruits 6 moves on the fruit line and arrives within the imaging range of the camera 2. At this time, the detector 3 is arranged to detect the pallet 1. (3) When the detector 3 detects the pallet 1, the processing device 4
inputs the images taken by the camera 2 into the image memory separately as a red image, a green image, or a blue image and latches them. (4) As shown in FIG. The image is divided into sub-regions each consisting of pixels (number of images), an evaluation value is calculated for each sub-region, and then the neural network 44 calculates the attributes of each sub-region (for example, normal color, flaw color, Determine the rust color, etc.). Examples of evaluation values include RGB average values and statistics in other color spaces. (5) Based on the determination results of the neural network 44, maps 50, 51, 52 of each defect are created as shown in FIG.
Create. (6) The size and number of each defect map 50, 51, 52 are determined from these defect maps 50, 51, 52, and the image processing device 45 makes a final determination.

[0016] In the above-described embodiment, the color tone discrimination of fruits was explained, but it goes without saying that the present invention can also be applied to detecting defects in industrial products that have defects that can be distinguished by differences in color.

[0017]

Effects of the Invention As described above with reference to the embodiments, according to the fruit defect detection device of the present invention, when determining a target to be processed of a new variety, several sample products for learning are used. Since the neural network is made to learn the evaluation value of each defect, it can be easily applied to various objects to be processed.

[Brief explanation of drawings]

FIG. 1 is an overall configuration diagram showing an embodiment of a fruit defect detection device of the present invention.

FIG. 2 is a configuration diagram of a processing device used in the fruit defect detection device of FIG. 1;

FIG. 3 is an explanatory diagram showing an example in which the fruit defect detection device of FIG. 1 is divided into sub-regions each composed of 2×2 pixels.

FIG. 4 is an explanatory diagram showing a state in which a captured image is divided in the fruit defect detection device of FIG. 1;

FIG. 5 is an explanatory diagram showing an example of a determination result of an image taken by the fruit defect detection device of FIG. 1;

[Explanation of symbols]

1 Palette 2. Camera 3 Detector 4 Processing equipment 7 Terminal 41 A/D converter 42 Image memory 44 Neural network 45 Image processing device 50 Distribution map

Claims (1)

[Claims] Claims: 1. Imaging means for outputting image signals obtained by imaging objects to be processed with different hues, and image information taken by the imaging means is divided into regions each consisting of a predetermined number of images, and each region is divided in advance. processing means for calculating a predetermined evaluation value and determining the presence or absence of a defect in the object to be processed based on the calculated value; the processing means includes an image memory for latching and storing the image information; an image processing unit that calculates the evaluation value based on information from a memory; and a neural network that determines to which defect the area belongs based on the evaluation value information from the image processing unit.
and means for creating a distribution map of each defect for each object to be processed based on the determination result of the neural network, and final determination of defects using the distribution map, wherein the neural network is used for processing objects to be processed for the first time. 1. A fruit defect detection device characterized in that, when determining an object, characteristics of the defective portion are collected as learning data using a plurality of types of objects to be processed for learning.