Object sorting system, object sorting method, and object sorting program

Abstract

To provide an object sorting system capable of sorting even new types of objects with high accuracy.SOLUTION: The system includes a photograph acquisition unit for acquiring a photograph of a captured object, a calculation unit for calculating the degree to which an object contained in the photograph indicates a type of waste, via a learning model pre-trained through machine learning on the correspondence between training image data and information indicating the types of objects contained in the training image data, and an output unit for outputting the degree together with the photograph.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to an object sorting system that acquires a photograph of waste, determines the type of waste, and generates a recycling score and comments. [Background technology]

[0002] The waste recycling process has traditionally been opaque, and with the rise in illegal dumping and waste disposal costs, and with carbon neutrality becoming increasingly important, there is a need for an appropriate waste sorting system.

[0003] The waste sorting system and method disclosed in Patent Document 1 includes a waste transport means, a photograph acquisition means, a display means, a material information determination means, and a system control means. The type of waste can be determined based on the material information contained in the waste.

[0004] The waste history management system disclosed in Patent Document 2 discloses a method for efficiently managing and analyzing waste data, and by managing waste history, it becomes possible to compare and analyze waste. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] Patent No. 5969685 [Patent Document 2] Japanese Patent Application Publication No. 2023-171971 Summary of the Invention [Problem to be solved by the invention]

[0006] In conventional object sorting systems, information indicating the type of object was pre-programmed, making it impossible to recognize the type of new object when it appeared.The present invention aims to provide an object sorting system that can sort even new types of objects with high accuracy by using machine learning to pre-learn the correspondence between information indicating the type of object and learning image data. [Means for solving the problem]

[0007] One embodiment of the present invention includes a photo acquisition unit that acquires a photograph of the captured object, a calculation unit that calculates the degree to which the object contained in the photograph indicates a type of waste through a learning model that has previously been machine-learned to determine the correspondence between training image data and information indicating the type of object contained in the training image data, and an output unit that outputs the degree together with the photograph.

[0008] The object classification system may further include a generation unit that generates degree-based evaluation data through a large-scale language model that has previously trained an input evaluation sentence text. The above-mentioned object sorting system may further include a price search unit that acquires images similar to the photograph as similar images and searches the Internet for the sales price range of similar images, and a conversion unit that converts the amount indicating the sales price range into points. The generating unit may generate an identification code that identifies the identity of the owner of the object and associate it with the photograph of the object.

[0009] This is an object sorting method that performs the following steps: a photo acquisition step in which a computer acquires a photograph of the captured object; a calculation step in which the computer calculates the degree to which the object contained in the photograph indicates a type of waste using a learning model that has previously machine-learned the correspondence between learning image data and information indicating the type of object contained in the learning image data; and an output step in which the computer outputs the degree together with the photograph.

[0010] This is an object sorting program that causes a computer to execute a photo acquisition function that acquires a photograph of an object that has been photographed, a calculation function that calculates the degree to which an object contained in a photograph indicates a type of waste through a learning model that has previously learned by machine learning the correspondence between learning image data and information indicating the type of object contained in the learning image data, and an output function that outputs the degree together with the photograph. [Effects of the Invention]

[0011] According to one embodiment of the present invention, the object sorting system of the present invention generates an information set including a photograph of the object, the degree to which the photographed object indicates a type of waste, an evaluation based on that degree, a score obtained by searching the sales price range of second-hand goods similar to the object, etc. Therefore, the present invention makes it possible to manage and analyze statistical information on waste through a large language model (LLM) in the waste sorting process. [Brief explanation of the drawings]

[0012] [Figure 1] 1 is a diagram showing an outline of an object sorting system according to an embodiment of the present invention. [Figure 2] 1 is a block diagram showing the configuration and functions of an object sorting system according to an embodiment of the present invention. [Figure 3] 10 is an example of an object sorting system according to an embodiment of the present invention generating a match rate and a comment. [Figure 4] 10 is an example of an object sorting system according to an embodiment of the present invention searching for the prices of similar items online and converting them into points. [Figure 5] 10 is an example of a separation result generated by the object separation system according to the embodiment of the present invention. [Figure 6] FIG. 10 is a flowchart illustrating an example of an operation of the object sorting system according to the embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0013] Hereinafter, an embodiment of the object sorting system 10 according to the present invention will be described with reference to the drawings. Note that the present invention is not limited to the embodiment. Also, in the drawings, some components that are not important for the explanation are omitted.

[0014] <Summary> 1, an object sorting system 10 according to the present invention is connected to a user terminal 20 via the Internet NW. The user terminal of this embodiment may be an electronic device such as a personal computer (hereinafter referred to as a PC), but functions may also be realized by any other electronic device other than a PC, such as a mobile phone, a smartphone, a tablet device, or a wearable device.

[0015] As shown in FIG. 2, the object sorting system 10 may be configured with a photo acquisition unit 101, a calculation unit 102, a generation unit 103, a price search unit 104, a conversion unit 105, and an output unit 106. However, the present invention is not limited to these units, and any unit may be configured to realize the functions of the object sorting system 10. The photo acquisition unit 101 of the present invention is configured to acquire a photo of an object. For example, if there is only one object within a bounding box, the bounding box used for photo acquisition in the object sorting system 10 may enclose the object to acquire a photo. If there are multiple objects within the bounding box, each bounding box may enclose the object and acquire an object image. Furthermore, the photo acquisition unit 101 acquires a photo of the object, and calculates the degree to which the object in the photo indicates a waste type through a learning model that has previously machine-learned the correspondence between training image data and information indicating the type of object contained in the training image data. The system then determines the product category using a database that defines the correspondence between product names and product categories, and outputs the degree, product category, and the photo together with the photo.

[0016] According to the present invention, the degree to which an object included in a photograph indicates a type of waste can be calculated through a learning model that has previously machine-learned the correspondence between training image data and information indicating the type of object included in the training image data. Furthermore, additional learning can be efficiently carried out by focusing additional learning on feature amounts that are particularly in need of additional learning.

[0017] The photo acquisition unit 101 may include an area division unit that performs processing to extract object areas from the image to be evaluated, and a processed feature extraction unit that extracts features from the divided object areas.The object classification system 10 can then perform processing to determine a learning model from the learning image data.With the above configuration, the photo acquisition unit 101 according to the present invention is characterized not only by high accuracy but also by high speed and energy saving.

[0018] The calculation unit 102 can calculate a match rate indicating whether an object in a photograph matches a predetermined waste type. For example, the object sorting system 10 can recognize an object in a photograph and then calculate the degree of similarity (hereinafter referred to as a match rate) with the predetermined type. The predetermined type may include waste types (e.g., circular economy, rare metal extraction, upcycling, composting, reuse, etc.). A method for recognizing objects in a photograph may be publicly known. The match rate, which indicates the correspondence between waste types and objects, may be calculated using a trained model. In this case, the trained model may be trained to learn the correspondence between feature values ​​generated based on the color of each pixel constituting a certain area in an image and whether the target waste is captured in that area. This correspondence is preset based on a sales performance database, allowing a determination of whether a photograph is waste.

[0019] Based on the degree calculated by the calculation unit 102, the generation unit 103 can generate information about the recycling of the object (hereinafter referred to as evaluation data), an identification code for identifying the identity of the owner of the object, a QR code (registered trademark), etc. The evaluation data is used to provide recycling guidance information to the user based on the matching rate. As an example, if a certain laptop computer has the highest matching rate with a predetermined type B (e.g., a reused product), the generation unit 103 may generate advice based on the matching rate, such as "This laptop computer can be recycled as a reused product, and the laptop computer screen can be used for other purposes."

[0020] For example, when the generation unit 103 connects to the Internet using a QR code (registered trademark) or the like as an information recording means (for example, when the current information is not directly recorded in the means attached to the QR code (registered trademark)), the generation unit 103 can use a processing device (CPU, etc.) and storage device such as a publicly known computer, server device, or mobile terminal as the recording and updating means.

[0021] The price search unit 104 can search for the price of a used item similar to the object by recognizing an object included in a photograph by the object sorting system 10. As an example, a certain discarded refrigerator may be sold at multiple secondhand stores for 1,000 yen, 2,000 yen, and 3,000 yen, and the price search unit 104 may generate a price range by collecting the prices of the secondhand stores. In this case, the price search unit 104 may set the price range to 1,000 to 3,000, or may calculate the median or average price to use as the search price. In addition to automatically searching for prices, the user may manually input the price of a used item into the price search unit 104 because the actual price of a used item fluctuates depending on the condition of the item and market conditions.

[0022] The price search unit 104 searches for records in the feature table where the average value of the feature included in the specified search criteria is equal to or greater than a predetermined value, and stores the results in the search result table of the search result database. The object sorting system 10 according to this embodiment uses color information, shape information, and the like as feature values. It may also be possible to extract and output only the product name, price excluding tax, price including tax, and barcode area from an e-commerce site, etc. The reference image, like the target image and non-target image, does not have to consist of only one image, but may also be a data set consisting of multiple images.

[0023] The conversion unit 105 can convert the prices into points based on the price range, median number, or average number of the prices of the price search unit 104. The conversion result is the same as the price search result, and the prices may be converted into points based on the median number or average number.

[0024] The output unit 106 can output information such as a photograph of the object, each identification code, and score to a user terminal. The output may be displayed on a smartphone app or via a medium such as a photograph. In addition, the output object recycling assessment sheet is tagged with each identification code, so anyone can check it.

[0025] The object sorting system 10 according to this embodiment is equipped with a generation AI including an LLM, and the generation AI's prompts may combine the process into one, or may be divided into multiple steps depending on the functions of the object sorting system 10. The prompts in this embodiment are divided into (1) sorting the objects, (2) investigating the used sales price range by searching for similar images, and (3) preventing tampering, but are not limited to these and may include steps such as editing EXIF ​​(Exchangeable image file format) and visualizing the recycling process.

[0026] <Sorting of objects> The object sorting system 10 according to this embodiment can be equipped with an image recognition model. Specifically, the learning model required for image recognition according to this embodiment is a model that has been machine-learned in advance to determine the correspondence between training image data and information indicating the type of object contained in the training image data, and the type of object contained in a photograph may be identified via the learned learning model.

[0027] Furthermore, the image identification technology is not limited to these, and an object contained in a photograph may be identified through an image detection model. As an example, the image identification technology is generated in advance by a so-called supervised machine learning method using learning data that associates an image area of ​​an object extracted from an infrared image captured in the infrared wavelength band with information indicating whether the object captured in this image area contains wavelength fluctuations.

[0028] The object sorting system 10 according to this embodiment can determine whether an object is waste from a photograph and output the result. Methods for acquiring the purpose include having the camera transmit the purpose and acquiring it, or having the user input it using the input / output unit of a computer.

[0029] The object sorting system 10 according to the present invention calculates the degree to which an object included in a photograph indicates a type of waste through a learning model that has previously learned the correspondence between learning image data and information indicating the type of object contained in the learning image data through machine learning. The object sorting system 10 can use a reference image (a reference image created using the method of the embodiment) obtained by converting a provisionally captured image, or a reference image (a reference image created using the method of the embodiment) obtained by converting a provisional reference image. The user can select either one.

[0030] On the other hand, the object sorting system 10 according to the present embodiment performs machine learning to determine the correspondence between information indicating the type of an identified object and information indicating the type of predetermined waste, and calculates the degree to which an object included in a photograph indicates a waste type. The method for analyzing the type of object is not limited, and information indicating the characteristics of the object in the photograph may be extracted and compared with information indicating the type of predetermined waste to determine whether or not there is a match. For example, the object sorting system 10 pre-extracts information such as the color, material, and shape of the object included in the photograph, assigns a value to it, and creates photo data. Meanwhile, the object sorting system 10 assigns a predetermined value to the information indicating the type of predetermined waste, creating predetermined type data. The calculation unit 102 then compares the photo data with the predetermined type data, and if the data falls within a certain range, determines that the object is highly related to the predetermined type. This method can improve the efficiency of waste sorting operations.

[0031] According to the present invention, it is possible to determine whether an object is waste from a photograph taken. The present invention is also applicable to mobile information terminals such as mobile phones and tablet terminals.

[0032] The present invention acquires a photograph of an object taken by a photo acquisition unit 101, and calculates the degree to which the object in the photograph indicates a type of waste through a learning model that has previously machine-learned the correspondence between learning image data and information indicating the type of object contained in the learning image data. In this process, more accurate determination is possible by taking into account not only color tone but also features such as shape and size. Furthermore, special waste such as heterogeneous waste and bulky waste can also be appropriately identified. Based on the degree calculated in this way, the output unit can sort the object into types such as general waste and recyclable waste.

[0033] The images input to the object classification system 10 according to the present invention may be assigned an identifier, which may be set in advance for the image or may be set by a similar image search device, so that the images can be identified.

[0034] The object sorting system 10 according to this embodiment can determine whether an object is waste from a captured image. Furthermore, since any image processing is possible, more accurate determination is possible. For example, the representative image may be the first image received from among the grouped images by the selection unit, or any image from among the grouped images.

[0035] The object classification system 10 according to the present invention generates a multimodal model by training a fully connected layer so that classification information that is likely to be output based on the appearance features and classification text features of training images approaches the classification information of the training images. This multimodal model receives the appearance features and classification text features of an input image and is adjusted so that the output classification information approaches the correct classification information of the input image. This multimodal model can then be used to estimate the correct classification information of the input image.

[0036] The object classification system 10 according to the present invention trains the attention mechanism of the appearance feature extraction model so that classification information likely to be output based on the internal state of the CNN unit, obtained by inputting appearance information of training images, approaches the classification information of the training images. In this way, the image similarity estimation system according to the present invention can estimate the degree to which a photograph containing the same type of object as the training image is similar to the training image by inputting the photograph. The image similarity estimation system according to the present invention can be used in object classification systems.

[0037] As shown in Fig. 2, the object sorting system 10 may acquire a photograph of an object 30, and then generate a list having a predetermined classification 301, a matching rate 302, and ratings and comments 303 based on the object 30 contained in the photograph. The number and definition of the predetermined classifications 301 are not particularly limited, and a user may set their own predetermined classifications or use classifications built into the object sorting system 10.

[0038] Furthermore, the object sorting system 10 according to the present invention extracts product regions from product images, calculates image features of each product, and calculates the similarity between the products. In response to a similar image search request input by a user, the object sorting system 10 can extract one or more highly similar products and present them to the user. The product recommendation system according to the present invention is expected to be used, for example, on online shopping sites on the Internet. The similar image search request input by a user may be, for example, a photograph taken by the user. Since the product recommendation system according to the present invention can extract one or more highly similar products in response to the similar image search request input by the user, the user can significantly reduce the time and effort required to search for the product they want.

[0039] In both the photo acquisition unit 101 and the price search unit 104, the calculation of the feature vector of the search target image does not necessarily have to be performed by the similar image search device of the present invention, and similar image search can be performed using the feature vector of the search target image calculated by another device.

[0040] The image retrieval method according to the present invention divides an image into regions with the same characteristics, performs pixel data matching for each divided region as a rough recognition using a reference template, extracts candidate objects, and detects the position of the objects. Then, to accurately detect the target object from the candidate objects, feature quantities such as the shape and color of the object are used. In this way, the image retrieval method according to the present invention enables high-precision, high-speed image retrieval.

[0041] In the object sorting system 10 according to this embodiment, the image generating device does not have to be a scanner, and may be a device capable of capturing images, such as a digital camera, a smartphone, or a tablet terminal.

[0042] Information on automatically obtainable photographing conditions, such as the camera model, may be automatically obtained and used to search for a reference image generation model. Also, the object to be inspected is not limited to food, but may be a package, an industrial product, etc.

[0043] <Investigating used car sales price ranges using similar image search> The price search unit 104 according to this embodiment may pre-register attribute information, such as identification information of combined product items associated as identical or similar, image feature data extracted from image data of the sales product items, sales information such as the price and brand of the sales product items, and usage information of each sales product, in a product item database. The price search unit 104 then compares an image captured by a user terminal with the product item database and, based on the comparison result, presents combined product items related to the product items photographed by the user. The product item database according to this embodiment is not limited to domestic online shops; it may also search for prices at overseas online shops, e-commerce sites, and brick-and-mortar stores. For overseas stores, the price search unit 104 may convert the foreign price into Japanese yen.

[0044] The object sorting system 10 according to this embodiment is also connected to the Internet NW and may search for products identical or similar to the object included in the photo from various secondhand goods sales websites. When an identical or similar product is found, attribute information, such as the photo, price, and seller, of the identical or similar product may be extracted and displayed on the user terminal 20. When performing a price search, a similar image search may be performed using a classifier that classifies used products. As shown in FIG. 4 , when a photo including an object 40 is input to the object sorting system 10, the object sorting system 10 may search for used or new products similar to the object 40 and generate a price list 401 containing information such as photos and prices of the used or new products. In this case, the information included in the price list 401 is not limited, and the price search unit 104 may add information such as the URL of an online store selling the used product and a product description to the price list 401. This information may be stored as category information in a database installed in the object sorting system 10.

[0045] The generation of the classifier does not necessarily have to be performed by the similar image retrieval device of the present invention, and similar image retrieval can be performed using a classifier generated by another device. It is also possible to automatically determine the type of photographed object and provide an appropriate processing method.

[0046] Category information is not only used as image feature information, but is also used in similarity calculations in the similar image search means, where the similarity between the images being compared is calculated by taking a weighted average of the similarities in color, shape, and texture. The category information may be stored in a database as text information. For example, the object sorting system 10 is connected to each e-commerce site, and the database includes product information. It may also store information stored in the storage unit, user input information, search results, and other information. If the user sets their own search conditions, the object sorting system 10 may also include a condition setting unit. In this case, the search results are narrowed down based on the conditions set by the user.

[0047] Furthermore, in the present invention, the IDs of the major category, the medium category, and the small category may be registered separated by underscores. For example, if the item to be recycled is a common everyday item such as a pin, lunch box, or plastic packaging, the item may be assigned a large category ID since it is an ordinary item. Conversely, if the item to be recycled is a valuable item with multiple parts such as a computer, tablet, or coffee maker, each part may be assigned a small category ID since the recycling method for each part differs depending on the material.

[0048] After obtaining the price of each used or new item, the conversion unit 105 may convert the price into a predetermined number of points according to a certain ratio. For example, the price search unit 104 may analyze information about the listing of each product and assign points to the object in a manner equivalent to the price of each used or new item. Referring to the second half of FIG. 4, the conversion unit 105 may enter the name 402 of a store selling the object, and then generate a list of information such as the price 403, photo 404, and point number 405 of an item similar to the object at that store.

[0049] When searching for used or new items similar to the target object, a learning model may be used to calculate the confidence level of the results. For example, if the confidence level of the judgment result by the trained model is 95% or higher, the object contained in the photo is determined to be highly related to the used or new item and a match, and the price of the used or new item is acquired. If the confidence level is less than 95%, the object contained in the photo is determined to be lowly related to the used or new item and a mismatch, and the price of the used or new item does not need to be acquired.

[0050] <Tamper prevention> The object sorting system 10 according to this embodiment may use Advanced Data Analysis to embed personal information of the object owner in an uploaded image using steganography. Steganography is a form of information hiding, a technique for embedding certain information within other information. As an example, the generation unit 103 of the object sorting system 10 according to this embodiment may generate a first identification code that reads the personal information of the object owner, or may generate a second identification code that reads information about the object. The first identification code and the second identification code are not limited to QR codes (registered trademark) and may include information identification codes such as barcodes.

[0051] According to FIG. 5, the object classification system 10 according to this embodiment may generate an information list indicating the analysis result of the object included in the photograph. The information list may have a photograph of the object, a first identification code for reading the personal information of the owner, and a second identification code for reading information about the object. Among them, the first identification code of the owner may simultaneously include information that can identify the identity of the owner. The information for identifying the identity of the owner may be input by the owner in advance into the database installed in the object classification system 10, or may be collected from publicly available information on the Internet and stored in the database. When the user reads the first identification code, the user can input the identification code on the screen presented when the user reads the first identification code. By comparing the identification code input by the user with the information that can identify the identity of the user stored in the database installed in the object classification system 10, if they match exactly, it can be determined that the user is the owner of the object. As an example, if the owner of the object is a school, the school may input the school code into the object classification system 10 in advance before obtaining the photograph of the object. After generating the recycling information of the object, the school can perform identity authentication by reading the first identification code and inputting the school code. If the school codes match, it may be determined that the owner of the object included in the photograph is the school.

[0052] After the user determines that they are the owner of the object included in the photograph, the user may read the second identification code and check the recycling information of the object. As an example, after the user reads the second identification code, product information, a proposal for the optimal recycling method, the estimated point number retrieved from similar images for the used sales price, the school code, EXIF information, etc. may be displayed on the user terminal 20.

[0053] <EXIF Editing> The object sorting system 10 according to this embodiment may link with other systems to add tags to the metadata of photographs. Generally, when information such as photographic data can be exchanged, there is a high possibility that information contained in the photographic data may be leaked due to EXIF ​​information. In this case, the object sorting system 10 displays EXIF ​​information such as the photograph's shooting date, tags, rating, size, title, creator, and shooting location on the user terminal 20. The user may re-edit each piece of information on the user terminal 20, or the object sorting system 10 may encrypt each piece of information. Furthermore, the EXIF ​​information may be re-edited using an EXIF ​​editing tool.

[0054] <Process Management> The object sorting system 10 according to this embodiment may also generate a status management screen and a reference dashboard to visualize the processes of collection by factories, reuse (upcycling) at schools, and other waste management activities. For example, information about a particular waste item, such as the waste disposal date, the time the recycler collects the waste, and the recycling results of the waste, may be displayed on the user terminal 20. The user can grasp the recycling information and recycling process for the waste item in real time.

[0055] The information processing device related to the object sorting system 10 of the present invention performs the learning process to generate a learning model, but this is not limited to this. A device different from the information processing device may generate the learning model, and the information processing device may obtain information on the learned learning model from this device, store it in the learning model memory section of the memory section, and use it for anomaly detection processing.

[0056] The photo acquisition unit 101 and price search unit 104 according to the present invention comprise an image acquisition module, an image display module, an area division module, an evaluation value calculation module, and an evaluation model determination module. The image acquisition module acquires the image to be evaluated. The image display module displays the image to be evaluated. The area division module divides the image to be evaluated into areas. The evaluation value calculation module calculates the feature amount of each area and calculates an evaluation value for each area. The evaluation model determination module calculates an evaluation value for the entire image to be evaluated from the evaluation values ​​of each area.

[0057] Furthermore, the implementation form of the program is not limited to an application program such as an object code compiled by a compiler or a program code executed by an interpreter, but may also be in the form of a program module incorporated into an operating system, etc. The code module may be stored in any type of computer-readable medium or other computer storage device.

[0058] The object sorting system 10 according to this embodiment can determine whether an object is waste from a photograph and propose an appropriate disposal method. Furthermore, machine learning improves accuracy, enabling more accurate sorting. The present invention can automatically determine the type of waste from a photograph. Furthermore, even if the direction of reclassification based on the reference data table differs, precise reclassification is possible.

[0059] <Operation> Next, with reference to FIG. 6, the operation of the object sorting system 10 according to this embodiment will be described. As shown in FIG. 5, the photo acquisition unit 101 acquires a photo of an object (S501) and then calculates the degree of similarity between the item in the photo and information indicating a predetermined type (S502). In this case, since multiple match rates are calculated for a certain object, the object sorting system 10 can generate an evaluation statement based on the match rates (S503). The price search unit 104 then searches the Internet for products similar to the object (S504) and calculates the value of the object. Next, the conversion unit 105 assigns a score to the object based on its value (S505). Then, the object sorting system 10 generates a list of information, including the object photo, identification code, and comment text, as output information (S506). Finally, the object sorting system 10 transmits the listed information to the user terminal 20 and displays it on the display unit of the user terminal (S507).

[0060] The object sorting system 10 according to the present invention is configured with electronic devices such as a personal computer, but may also be implemented with any other electronic devices, such as a mobile phone, smartphone, tablet device, or wearable device, in addition to a PC. The present invention makes it possible to determine whether an object is waste based on a photograph. The user terminal 20 itself may also be equipped with an AI edge device. The AI ​​edge device is the AI ​​portion of an IoT device equipped with artificial intelligence (AI).

[0061] Some of the CPU functions may be realized by circuits executed by program code such as a DSP, or by hardware configurations such as gate circuits generated based on a programming language written in Verilog, or by hardware circuits.

[0062] The program may be stored in advance in a storage device (a storage device with a non-transitory storage medium) such as an HDD (Hard Disk Drive) or flash memory, or may be stored in a removable storage medium (a non-transitory storage medium) such as a DVD or CD-ROM and installed by inserting the storage medium into a drive device. The program may be for realizing some of the functions described above, or may be capable of realizing the functions described above in combination with a program already stored in the computer system, or may be realized using a programmable logic device such as an FPGA (Field Programmable Gate Array).

[0063] The object sorting system 10 according to this embodiment can determine whether an object is waste from a captured image, thereby improving the efficiency of waste recycling work. In addition, the use of artificial intelligence technology such as LLM makes it possible to obtain useful recycling comments for each object, which is expected to lead to more efficient recycling work. [Explanation of symbols]

[0064] 10. Object sorting system 20 User terminal 30 Examples of Objects 101 Photo Acquisition Department 102 Calculation Unit 103 Generation part 104 Price Search Section 105 Conversion Department 106 Output section

Claims

1. a photo acquisition unit that acquires a photo of the captured object; a calculation unit that calculates the degree to which the object included in the photograph indicates a type of waste through a learning model that has previously been machine-learned to learn a correspondence between learning image data and information indicating the type of object included in the learning image data; an output unit that outputs the degree together with the photograph. Object sorting system.

2. The object sorting system includes: a generation unit that generates evaluation data based on the degree through a large-scale language model that has previously trained the input evaluation sentence text.

2. The object sorting system according to claim 1.

3. The object sorting system is a price search unit that acquires an image similar to the photograph as a similar image and searches the Internet for a sales price range of an object included in the similar image; a conversion unit that converts the amount indicating the sales price range into points.

2. The object sorting system according to claim 1.

4. The generation unit An identification code that identifies the identity of the owner of the object is generated and linked to the photograph of the object.

3. The object sorting system according to claim 2.

5. The computer a photo acquisition step of acquiring a photo of the captured object; a calculation step of calculating the degree to which the object included in the photograph indicates a type of waste through a learning model that has previously been machine-learned to learn the correspondence between learning image data and information indicating the type of object included in the learning image data; and an output step of outputting the degree together with the photograph. Method of separating objects.

6. On the computer, a photo acquisition function for acquiring a photo of the captured object; a calculation function that calculates the degree to which the object included in the photograph indicates a type of waste through a learning model that has previously been machine-learned to learn the correspondence between learning image data and information indicating the type of object included in the learning image data; and and an output function of outputting the degree together with the photograph. Object sorting program.

Images