Device and method for classifying fruits
Disclosed is a device and method for identifying fruits that sorts fruits by sequentially sorting a plurality of fruits which are input, generating image data, determining the grade of each fruit based on the weight of the fruit, and determining a defective state or a normal state by using a defect feature certainty.
1 . A fruit identification device comprising:
a fruit inserting unit configured to align a plurality of fruits in a row and to rotate the aligned fruits;
a fruit sorting unit configured to sequentially classify the aligned fruits under control of a control unit;
an image acquisition unit configured to generate pieces of image data for one fruit by sequentially capturing the rotated fruits; and
wherein the control unit configured to:
generate pieces of fruit image data by extracting only a fruit from the pieces of image data;
segment each piece of fruit image data into a plurality of localized image fragments corresponding to different surface regions of the fruit;
determine, using an artificial-intelligence model, a defect probability value for each of the localized image fragments;
aggregate the defect probability values of the localized image fragments to generate a plurality of defect feature certainties for the fruit;
compare the defect feature certainties with a predetermined reference defect feature certainty;
determine that the fruit is in a defective state or a normal state based on the comparison result;
determine a weight of the fruit by using the pieces of fruit image data;
determine a grade of the fruit by using the weight of the fruit;
when the fruit is in the defective state, move the fruit to a defective region through the fruit sorting unit, and generate a visual projection of defect regions onto the localized image fragments for display; and
when the fruit is in the normal state, move the fruit to a normal region corresponding to the grade of the fruit among a plurality of normal regions through the fruit sorting unit.
2 . The fruit identification device of claim 1 , wherein the control unit is configured to:
when at least one of the defect feature certainties is greater than or equal to the reference defect feature certainty, determine that the fruit is in the defective state; and
when all of the defect feature certainties are less than the reference defect feature certainty, determine that the fruit is in the normal state.
3 . The fruit identification device of claim 1 , wherein the defective state includes at least one of a wound, a bruise, a spoilage, or pest infestation of a fruit.
4 . The fruit identification device of claim 1 , further comprising:
a display unit,
wherein the control unit is configured to:
when the fruit is in the defective state, display the grade of the fruit, the defect feature certainties, and the pieces of fruit image data, onto which a defect feature is projected, through the display unit; and
when the fruit is in the normal state, display the grade of the fruit and the defect feature certainties through the display unit.
5 . The fruit identification device of claim 1 , further comprising:
a communication unit,
wherein the control unit is configured to:
generate identification result data for the fruits; and
transmit the identification result data to a mobile terminal of a user through the communication unit.
6 . A method for identifying a fruit, the method comprising:
aligning, by a control unit, a plurality of fruits in a row and rotating the aligned fruits, through a fruit inserting unit;
generating, by the control unit, pieces of image data for one fruit by sequentially capturing the rotated fruits through an image acquisition unit;
generating, by the control unit, pieces of fruit image data by extracting only a fruit from the pieces of image data;
segmenting, by the control unit, each piece of fruit image data into a plurality of localized image fragments corresponding to different surface regions of the fruit;
determining, by the control unit and using an artificial-intelligence model, a defect probability value for each of the localized image fragments;
aggregate, by the control unit, the defect probability values of the localized image fragments to generate a plurality of defect feature certainties for the fruit;
compare, by the control unit, the defect feature certainties with a predetermined reference defect feature certainty;
determine, by the control unit, that the fruit is in a defective state or a normal state based on the comparison result;
determining, by the control unit, a weight of the fruit by using the pieces of fruit image data and determining a grade of the fruit by using the weight of the fruit;
when the fruit is in the defective state, moving, by the control unit, the fruit to a defective region, and generating a visual projection of defect regions onto the localized image fragments for display; and
when the fruit is in the normal state, moving, by the control unit, the fruit to a normal region corresponding to the grade of the fruit among a plurality of normal regions.
7 . The method of claim 6 , wherein the determining that the fruit is in the defective state or the normal state includes:
when at least one of the defect feature certainties is greater than or equal to the reference defect feature certainty, determining that the fruit is in the defective state; and
when all of the defect feature certainties are less than the reference defect feature certainty, determining that the fruit is in the normal state.
8 . The method of claim 6 , wherein the defective state includes at least one of a wound, a bruise, a spoilage, or pest infestation of a fruit.
9 . The method of claim 6 , further comprising:
generating, by the control unit, identification result data for the fruits; and
transmitting, by the control unit, the identification result data to a mobile terminal of a user through a communication unit.
10 . The fruit identification device of claim 1 ,
wherein the artificial-intelligence model comprises a convolutional neural network including a plurality of convolution layers, pooling layers, and fully connected layers configured to output the defect probability values for the localized image fragments.
11 . The fruit identification device of claim 1 ,
wherein determining that the fruit is in the defective state comprises determining that at least one of the defect feature certainties is equal to or greater than 90 percent (90%).
12 . The fruit identification device of claim 1 ,
wherein the control unit is further configured to generate a visual projection of defect regions onto corresponding localized image fragments for display.