TWI808845B - Fresh tea leaf grading method and device thereof - Google Patents

Abstract

The present invention discloses a fresh tea leaf grading method. With capturing a batch fresh tea image of a plurality of single fresh tea leaves; performing object detection operation to generate a plurality of single fresh tea information corresponding to the plurality of single fresh tea leaves; performing segmentation operation to each of the plurality of single fresh tea information to generate a fresh tea portion information; judging the fresh tea portion information according to a single fresh tea grade criteria to generate a single fresh tea grade information corresponding to each of the plurality of single fresh tea leaves. As a result, in conventional art, human identification errors and time spent can be reduced, and the single fresh tea grade criteria can be adjusted according to different tea species, seasons, origins and other factors without repeated use of human identification.

Description

Method and device for classifying tea cyanine

The invention relates to a method for identifying the grade of cyanine and its device, in particular to an identification method used in the technical field of commercial cyanine, which is used to improve the existing identification method.

In the field of commercial tea cyanine, the prior art uses manual grade identification, which requires a lot of time and manpower.

The current common methods of judging the quality of tea greens rely mostly on human observation. The criteria for judging include the length (uniformity), color (old and tender), length of internodes (growth conditions or seasons), leaf stem ratio (growth conditions or seasons), etc. of harvested tea greens. However, human judgment is highly dependent on long-term accumulation of experience, so the human resources for judgment are limited, and subjective disputes inevitably arise, and there is a lack of objective standards.

In recent years, due to the rise of the tea beverage culture, domestic demand for commercial tea greens has increased significantly. The average amount of tea consumed by Chinese people has risen from 0.27 kg in 1976 to 1.5 kg in 2017. However, at the end of 2019, due to the unstable quality of commercially picked tea extracts in northern tea regions, this caused doubts on the purchasing side, and the problem of unsalable tea extracts broke out for the first time.

Therefore, if the picking standard for commercial tea greens can be established to help improve the quality of commercial tea greens, it will be beneficial for tea farmers and tea factories to reach a consensus on the sales price. tea leaf picking The quality of tea affects the quality of subsequent tea production. If grading can be carried out at the tea green stage, the tea factory can adjust the tea production conditions and give full play to the quality of tea green, and at the same time save the cost of subsequent refined processing.

Therefore, there are technical problems in the known technology: 1. The grade identification of tea cyanine relies on human observation; 2. There is no objective data to adjust the classification basis arbitrarily and in real time.

Therefore, it is necessary to propose a method and device for identifying the grade of cyanine to solve the above-mentioned technical problems.

In order to solve the above-mentioned problems of the conventional technology, the present invention provides a cyanine grade identification system, which uses artificial intelligence to perform image processing on a batch of cyanine images to obtain multiple single cyanine information and cyanine part information corresponding to multiple single cyanine; then judge the cyanine part information to confirm the grade of each cyanine and evaluate the grade of the batch of cyanine.

In order to achieve the above object, the present invention provides a method for classifying cyanine, which includes: firstly, extracting a batch of cyanine images including a batch of cyanine, wherein the batch of cyanine contains a plurality of single cyanine; then, performing an object detection operation on the batch of cyanine images to generate a plurality of single cyanine information corresponding to the plurality of single cyanine; then, performing a block division operation on each of the plurality of single cyanine information to generate a cyanine part information; Judging to generate a single cyanine level message corresponding to each of the plurality of single cyanines.

In a preferred embodiment, the method further includes: firstly, capturing at least one training image; then, using the at least one training image to perform a deep learning calculation on the single cyanine part; and then, calculating the single cyanine level criterion.

In a preferred embodiment, the block division operation adopts a convolutional neural network.

In a preferred embodiment, the method further includes: fusing the single cyanine grade information of each of the plurality of single cyanines to generate a batch of cyanine information; then, judging the batch of cyanine information according to a batch of cyanine grade criteria to generate a batch of cyanine grade information corresponding to the batch of cyanine.

In a preferred embodiment, the method further includes: adjusting the single cyanine grade criterion and/or the batch cyanine grade criterion according to at least one reference message.

In a preferred embodiment, the method further comprises: placing the batch of cyanine on at least one scale template.

In a preferred embodiment, the method further comprises: making gaps exist between the plurality of single cyanines.

To achieve the above purpose, the present invention also provides a cyanine level identification device. The device utilizes the method described above. It includes: an image capturing unit, an image processing unit and a grade judging unit. The image capture unit is used to capture the batch of cyanine images. The image processing unit is used to electrically connect with the image capture unit and process the batch of cyanine images to obtain the plurality of single cyanine information and the cyanine part information. The grade judging unit is used for electrically connecting with the image processing unit and generating the grade information of the single cyanine.

In a preferred embodiment, the device further includes an actuating unit, which is used to electrically connect with the image processing unit and cooperate with the image capturing unit to make a gap exist between the plurality of single cyanines.

In a preferred embodiment, the image capture unit, the image processing unit, and the grade judging unit are compiled into at least one application program in the form of a library, variables or operands, and then built in a microprocessor of the cyanine grade identification device 100 .

Compared with the conventional technology, the present invention uses artificial intelligence to perform image processing on a batch of cyanine images to obtain multiple single cyanine information and cyanine part information corresponding to multiple single cyanine; then judge the cyanine part information to confirm the grade of each cyanine and evaluate the grade of the batch of cyanine.

100: Cyanine level identification device

110: image capture unit

120: Image processing unit

130: Level Judgment Unit

140: Action unit

S01-S09: Steps

Fig. 1 shows the first flow chart of the cyanine level identification method according to the present invention; Fig. 2 shows the second flow chart of the cyanine level identification method according to the present invention; Fig. 3 shows the third flow chart of the cyanine level identification method according to the present invention; Fig. 4 shows the artificial judgment criterion used in deep learning calculation according to the present invention;

Fig. 6 shows the block segmentation results of a single cyanine message in Fig. 4; Fig. 7 shows the actual operation diagram A of the application program of the cyanine grade identification method according to the present invention (single cyanine/batch cyanine grade identification result)

Fig. 8 shows the actual operation diagram B of the application program of the cyanine grade identification method according to the present invention (single cyanine/batch cyanine grade criterion); and Fig. 9 shows a schematic diagram of the cyanine grade identification device of the present invention.

The following descriptions of the various embodiments refer to the drawings to illustrate specific embodiments in which the present invention can be implemented. The directional terms mentioned in the present invention, such as "upper", "lower", "front", "rear", "left", "right", "inner", "outer", "side", etc., are only referring to the directions of the drawings. Therefore, the directional terms used are used to illustrate and understand the present invention, but not to limit the present invention.

Deep learning is an algorithm based on representational learning of data in machine learning. Observations (such as an image) can be represented in a variety of ways, such as a vector of intensity values for each pixel, or more abstractly as a series of edges, regions of a specific shape, etc. Whereas it is easier to learn tasks from examples (e.g., face recognition or facial expression recognition) using certain representations. The advantage of deep learning is to use unsupervised or semi-supervised feature learning and hierarchical feature extraction efficient algorithms to replace manual feature acquisition.

The goal of representation learning is to seek better representations and build better models to learn these representations from large-scale unlabeled data. Representation methods come from neuroscience, and are loosely based on information processing in similar nervous systems and an understanding of communication patterns, such as neural codes, which attempt to define the relationship between the responses that pull neurons and the relationship between the electrical activity of neurons in the brain.

So far, there have been several deep learning frameworks, such as deep neural network, convolutional neural network, deep belief network and loop neural network, which have been applied in the fields of computer vision, speech recognition, natural language processing, audio recognition and bioinformatics, and have achieved great results. Good results.

Please refer to FIG. 1 , which shows a flow chart of the method for identifying the grade of cyanine according to the present invention. Firstly, step S01 is executed to capture a batch of cyanine images including a batch of cyanine, wherein the batch of cyanine contains a plurality of single cyanine; then, step S02 is executed to perform object detection (Object Detection) operation on the batch of cyanine images to generate a plurality of single cyanine messages respectively corresponding to the plurality of single cyanine; then, step S03 is executed to perform block segmentation (Segmentation) operation on each of the plurality of single cyanine information to generate a cyanine part message; then, execute step S 04. Judging the cyanine part information according to a single cyanine level criterion to generate a single cyanine level information corresponding to each of the plurality of single cyanine levels. Through the cyanine level identification method of the present invention, a plurality of cyanine information (i.e. images) can be obtained through block segmentation from an image with multiple cyanine, and then each single cyanine information is segmented into blocks to obtain a cyanine part information (i.e. Mature Leaf, Old Leaf, Bud, Fresh Leaf, Fish Leaf, Red Stalk and Internode Length in a single cyanine). Finally, a single cyanine grade information is generated by evaluating the single cyanine through a single cyanine grade criterion. For example, a single cyanine grade can be divided into three grades A-C (A is the best, C is the worst).

Specifically, the single cyanine information and the cyanine part information are automatically operated by artificial intelligence. Preferably, a convolutional neural network (Convolutional Neural Network, CNN) can be used, or as in the present invention, a fast region-based convolutional neural network (Faster Region-based Convolutional Neural Networks, Faster R-CNN) and a mask region convolutional neural network (Mask Region-based based Convolutional Neural Networks, Mask R-CNN) as an algorithm for deep learning (Deep Learning).

A CNN is a specialized feed-forward neural network for processing data with a known, grid-like topology, such as image data. Therefore, CNNs are often used in computational vision and image recognition applications. Nodes in the input layer of a CNN are organized as a set of "filters" (feature detectors inspired by receptive fields found in the retina), and the output of each set of filters is passed to nodes in subsequent layers of the network. Computation for CNNs involves applying convolution mathematical operations to each filter to produce that filter's output. Convolution is a specialized mathematical operation performed by two functions to produce a third function that is a modified version of one of the two original functions. In convolutional network terminology, the first function of convolution can be called input, and the second function can be called convolution kernel (Convolution Kernel). This output may be referred to as a Feature Map. For example, the input to a convolutional layer may be a multidimensional matrix of data defining the various color components of the input image. The convolution kernel can be a multi-dimensional matrix of parameters, where the parameters are adjusted through the training process of the neural network.

It should be noted that although the convolutional neural network is used as the main deep learning basis in the present invention, it does not limit the use of other artificial intelligence tools for deep learning.

It should be noted that the cyanine treated in the present invention is unprocessed fresh tea leaves.

Fig. 2 shows the second flowchart of the method for identifying the grade of cyanine according to the present invention. The difference between this flow chart and the first flow chart is as follows: after step S04, then Next, step S05 is executed to generate a batch of cyanine information by fusing the single cyanine grade information of each of the plurality of single cyanine; then, step S06 is executed to judge the batch of cyanine information according to a batch of cyanine grade criteria to generate a batch of cyanine grade information corresponding to the batch of cyanine. A batch of cyanine information is generated from the plurality of single cyanine grade information obtained in step S04 , and then a batch of cyanine grade information is generated by evaluating the batch of cyanine grade information through a batch of cyanine grade information. For example, the grade of a batch of cyanine can be divided into grades 1-3 (grade 1 is the best, grade 3 is the worst).

Please refer to Figure 3. Fig. 3 shows the third flow chart of the cyanine level identification method according to the present invention; the difference between this flow chart and the second flow chart is as follows: step S07 is executed sequentially before step S01, and at least one training image is captured; then, step S08 is executed, and the deep learning calculation of the single cyanine part is performed with the at least one training image; then, step S09 is executed, and the single cyanine level criterion is calculated. The learning method is as follows: define the single cyanine part of the at least one training image, please refer to Figure 4, and make A-C judgments based on red stems, fish leaves, length, number of leaves, and whether they are all old leaves. After the above-mentioned training, the single cyanine level criterion will be obtained.

In detail, a total of 915 images (including 733 images for training and 182 images for testing) are used in the present invention, please refer to Table 1. In 733 images for training, 3718 single cyanines were defined as grade A, 6434 single cyanines were defined as grade B, and 6215 single cyanines were defined as grade C; in 182 images for testing, 1081 single cyanines were defined as grade A, 1294 single cyanines were defined as grade B, and 1623 single cyanines were defined as grade C.

Further, in the training method here, seasons (spring, summer, June white, autumn, winter), region, altitude, cyanine species, and even cyanine species from different farms can be used as benchmarks for different single cyanine level criteria. After the standard is established, the single cyanine part is defined on the at least one training image, and a deep learning model can be established to obtain the single cyanine grade/batch of cyanine grade.

It should be noted that in this application, the single cyanine level criterion/the batch cyanine level criterion is generated by using semi-supervised learning and supervised learning.

FIG. 4 illustrates human judgment criteria used in deep learning algorithms according to the present invention. Referring to Figures 1-3 at the same time, under certain circumstances (harvest/bad harvest/spring tea/winter tea, etc.), it may be necessary to increase/decrease the grading standard, and a step can also be performed: adjusting the single cyanine grade criterion and/or the batch of cyanine grade information according to at least one reference information. In this way, a certain amount of A-grade single cyanine or a first-class batch of cyanine can be maintained during harvest/poor harvest. The single cyanine can be manually re-graded by adjusting the single cyanine grade criterion and/or the batch cyanine grade information. In other words, the at least one reference information includes tea type, season, region, and even yield, which can be used as a basis for adjusting the single cyanine grade criterion and/or the batch of cyanine grade information.

In detail, the single cyanine grading criterion is manually defined, for example, the standard shown in Figure 4 is a single cyanine grading standard defined by experts and scholars.

Preferably, the step of placing the batch of cyanine on at least one scale template can also be performed. In this way, the size of the cyanine can be confirmed more easily.

Preferably, a step may also be performed such that gaps exist between the plurality of single cyanines. In this way, the object detection operation can be made more accurate.

FIG. 5 shows the original images of a batch of cyanine images and block images corresponding to a plurality of single cyanine information. The green tea here is Taicha No. 12 harvested by Guanghui Tea Factory (Oolong Tea Farm) on March 27, 2021 in the Republic of China. It can be known that after the object detection operation, a single cyanine is selected with a box and after the block division operation, its single cyanine grade information is marked beside it (A-C).

Please refer to FIG. 6 , which shows a block segmentation result diagram of a single cyanine message in FIG. 4 . By inputting the user (tea factory), season, tea species, and date on the left side, you can obtain a single cyanine level criterion and/or a batch of cyanine level information, and then obtain the information on the cyanine part on the right (heart bud: 1; young leaf: 1; mature leaf: 5; old leaf: 0; fish leaf: 0; red stem: 1; internode length: 4.66cm).

Fig. 7 shows the actual operation diagram A of the application program of the cyanine grade identification method according to the present invention (identification result of a single cyanine/batch cyanine grade). The difference from FIG. 6 is that FIG. 7 also includes a batch of cyanine information and a batch of cyanine grade information. This batch of cyanine information includes 4 pieces of A grade, 5 pieces of B grade, 20 pieces of C grade, and a total of 29 single cyanine pieces. The grade information of this batch of cyanine shows that this batch of cyanine is grade 3.

The present invention uses image processing combined with deep learning algorithms to bring technical effects that cannot be achieved by the prior art. Using the above-mentioned convolutional neural network to perform object detection on each batch of cyanine images, and separate the images of each single cyanine; then perform block segmentation to separate different parts of a single cyanine (ie, the above-mentioned cyanine part information). Of course, the object detection and block segmentation operations can be performed by the computer itself; preferably, the present invention uses the at least one training image to train the object detection and block segmentation operations, which can greatly improve the judgment result.

Referring to FIG. 8 , it shows the actual operation diagram B of the application program of the cyanine grade identification method according to the present invention (single cyanine/batch cyanine grade criterion). The single cyanine grade criterion is on the left, and the single cyanine grade criterion can be adjusted for any batch of cyanine according to different situations. The right side is the batch cyanine level criterion, and the batch cyanine level criterion can be adjusted for any batch of cyanine according to different situations. It should be noted that the parameter constraints in FIG. 8 and FIG. 4 can be adjusted according to the at least one reference information to adjust the single cyanine grade criterion and/or the batch cyanine grade criterion

Please refer to FIG. 9 , which is a schematic diagram of a cyanine level identification device 100 of the present invention. Please refer to FIGS. 1-8 at the same time. The cyanine level identification device 100 is used to implement the above-mentioned cyanine level identification method, which includes an image capture unit 110 , an image processing unit 120 and a level determination unit 130 . Among Fig. 9, put a batch of tea greens on a conveyer belt and carry out continuous identification operation. The image capture unit 110 is used to capture the batch of cyanine images. The image processing unit 120 is used to electrically connect with the image capture unit 110 and process the batch of cyanine images to obtain the plurality of single cyanine information and the cyanine part information. The level judging unit 130 is used to electrically connect with the image processing unit 120 and generate The single cyanine grade information. Preferably, after calculating the batch of cyanine images to obtain the single cyanine grade information of the plurality of single cyanines, the grade judging unit 130 may also fuse the single cyanine grade information of the plurality of single cyanines to obtain a batch of cyanine information, finally. The batch of cyanine information is judged by a batch of cyanine grade criteria to generate a batch of cyanine grade information corresponding to the batch of cyanine.

It should be noted that the image capturing unit 110 , the image processing unit 120 and the grade judging unit 130 are compiled into at least one application program in the form of a library, variables or operands, and then built in a microprocessor of the cyanine grade identification device 100 .

Preferably, the cyanine grade identification device 100 further includes an actuating unit 140 for electrically connecting with the image processing unit 120 and cooperating with the image capturing unit 120 to make gaps exist between the plurality of single cyanines. The actuating unit 140 may be a vibrating element or other mechanical components. Misjudgment by the image processing unit 120 is avoided by vibrating the conveyer belt or using other mechanical means to make a gap (a plurality of single cyanines are regarded as one).

Compared with the conventional technology, the present invention uses artificial intelligence to perform image processing on a batch of cyanine images to obtain multiple single cyanine information and cyanine part information corresponding to multiple single cyanine; then judge the cyanine part information to confirm the grade of each cyanine and evaluate the grade of the batch of cyanine.

The above are only preferred embodiments of the present invention. It should be pointed out that for those skilled in the art, without departing from the principles of the present invention, some improvements and modifications can also be made, and these improvements and modifications should also be considered as protection scope of the present invention.

S01-S04: Steps

Claims (10)

A method for identifying cyanine levels using a computer system, comprising: capturing a batch of cyanine images containing a batch of cyanine, wherein the batch of cyanine contains a plurality of single cyanine; performing an object detection operation on the batch of cyanine images to generate a plurality of single cyanine information corresponding to the plurality of single cyanine; performing a block division operation on each of the plurality of single cyanine information to generate a cyanine part information; A single cyanine level information; wherein the cyanine part information includes the number of mature leaves, the number of old leaves, the number of heart buds, the number of young leaves, the number of fish leaves, the presence or absence of red stems and the length of internodes of each of the plurality of single cyanines. The method for identifying cyanine levels according to claim 1, further comprising: capturing at least one training image; performing cyanine deep learning calculations with the at least one training image; and calculating the single cyanine level criterion. The method for discriminating cyanine grades as claimed in claim 1, wherein the block segmentation operation adopts a convolutional neural network. The method for identifying the grade of cyanine according to claim 1, further comprising: fusing the single cyanine grade information of each of the plurality of single cyanine to generate a batch of cyanine information; and judging the batch of cyanine information based on a batch of cyanine grade criteria to generate a batch of cyanine grade information corresponding to the batch of cyanine. The method for identifying cyanine grades according to claim 4, further comprising: adjusting the single cyanine grade criterion and/or the batch cyanine grade criterion according to at least one reference message. The method for identifying the grade of cyanine according to claim 1, further comprising: placing the batch of cyanine on at least one scale template. The method for identifying the grade of cyanine according to claim 1, further comprising: making gaps exist between the plurality of single cyanine. A cyanine level identification device, which uses the cyanine level identification method as described in any one of claims 1-7, comprising: an image capture unit, used to capture the batch of cyanine images; an image processing unit, electrically connected to the image capture unit and used to process the batch of cyanine images and obtain the plurality of single cyanine information and the information of the cyanine part; a grade judgment unit, electrically connected to the image processing unit and used to generate the single cyanine level information; The number of mature leaves, the number of old leaves, the number of heart buds, the number of young leaves, the number of fish leaves, the presence or absence of red stems and the length of internodes of a single tea cyanine. The cyanine grade identification device as claimed in claim 8 further comprises an actuating unit, which is used to electrically connect with the image processing unit and cooperate with the image capturing unit to make a gap exist between the plurality of single cyanines. The cyanine level identification device as described in claim 9, wherein the image capture unit, the image processing unit and the level judgment unit are compiled into at least one application program in the form of a library, variable or operator, and then built in a microprocessor of the cyanine level identification device.