Training of vision detection systems using RFID tags

Abstract

An item identification system includes a resolver configured to resolve item classes from item identifiers provided by radio frequency identification (RFID) tags and a machine learning system that is trained to learn associations between data about items, such as images of items, and resolved item classes. When an item is to be identified, data about the item can be captured and routed to the learning system. The learning system may be sufficiently trained such that it is able to identify the item, even if the item does not have an associated RFID tag to provide item class information. If the item does have an associated RFID tag, an item class of the item may be resolved from information provided by the RFID tag and is used along with the captured data about the item to further train the learning system.

Description

TRAINING OF VISION DETECTION SYSTEMS USING RFID TAGSBACKGROUND OF THE INVENTION

[0001] Radio-Frequency identification (RFID) systems typically include RFID readers, also known as RFID reader / writers or RFID interrogators, and RFID tags, also known as radio tags. RFID systems can be used in many ways for locating and identifying objects to which the tags are attached. RFID systems are useful in product-related and service- related industries for tracking objects being processed, inventoried, or handled. In such cases, an RFID tag is usually attached to an individual Item, or to its package, The RFID tag:typically includes, or is, a radio-frequency (RF) Integrated circuit (IC). Such an RFIC may be referred to as an RFID IC or a radio IC.

[0002] In principle, RFID techniques entail using an RFID reader to inventory one or more RFID tags, where inventorying involves singulating a tag, receiving an identifier from a tag, and / or acknowledging a received identifier (e.g., by transmiting an acknowledge command). " Singulated" is defined as a reader singling-out one tag, potentially from among multiple tags, for a reader-tag dialog, " Identifier" is defined as a number or value identifying the tag or the item to which the tag is attached, such as a tag identifier (TID), electronic product code (EPC), etc. An ''inventory round" is defined as a reader staging RFID tags for successive inventorying. The reader transmitting an RF wave performs the Inventory. The RF wave Is typically electromagnetic, at least in the far field. The RF wave can also be predominantly electric or magnetic in the near or transitional near field. The RF wave may encode one or more commands that instruct the tags to perform one or more actions. The operation of an RFID reader sending commands to an RFID tag is sometimes known as the reader "interrogating" the tag.

[0003] In typical RFID systems, an RFID reader transmits a modulated RF inventory signal (a command), receives a tag reply, and transmits an RF acknowledgement signal responsive to the tag reply. A tag that replies to the interrogating RF wave does so by transmitting back another RF wave. The tag either generates the transmitted back RFwave originally, or by reflecting back a portion of the interrogating RF wave in a process known as backscatter. Backscatter may take place in a number of ways.

[0004] The reflected-back RF wave, may encode data stored in the tag, such as a number. The response is demodulated and decoded by the reader, which thereby identifies, counts, or otherwise interacts with the associated item. The decoded data can denote a serial number, a price, a date, a time, a destination, an encrypted message, an electronic signature, other attribute(s), any combination of attributes, and so on. Accordingly, when a reader receives tag data it can learn about the Item that hosts the tag and / or about the tag itself.[0005 An RFID tag typically includes an antenna section, a radio section, a power¬ management section, and frequently a logical section, a memory, or both. In some RFID tags the power-management section includes an energy storage device such as a battery. RFID tags with an energy storage device are known as battery-assisted, semi¬ active, or active tags. Other RFID tags can be powered solely by the RF signal they receive. Such RFID tags do not include an energy storage device and are called passive tags. Of course, even passive tags typically include temporary energy- and data / flag- storage elements such as capacitors or inductors.SUMMARY OF THE INVENTION

[0006] This summary is provided to introduce a selection of concepts in a simplified form that are further described below in the Detailed Description. This summary is not intended to identify key features or essential features of the claimed subject matter, nor is it intended as an aid in determining the scope of the claimed subject matter.

[0007] Some examples are directed to a method for an interface to facilitate item identification. The method may include receiving a first Item class, in one example, the first item class is obtained from a resolver configured to resolve an item class from an item identifier, the item identifier originating from a radio frequency identification (RFID) integrated circuit (IC) associated with an item. In another example, the first item class Is obtained from a learning system configured to receive an image-class pair, the image-class pair comprising an image of an item and the resolved item class, and to usethe image-class pair and a training algorithm to learn an association between the item image and resolved item class. The method also includes determining a destination for the first item class and transmitting the first item class to the destination.

[0008] Other examples are directed to a computing device configu ed to execute an item information system. The computing device includes a communication subsystem configured to communicate with an item identification system, a memory to store instructions, and a processor coupled to the communication subsystem and the memory. The processor can be configured to execute an interface in conjunction with the instructions stored in the memory, and be configured to receive, via the interface, the first item class. The processor may additionally be configured to determine a destination for the first item class, and to transmit the first item class to the destination. The item identification system may include a resolver configured to resolve an item class from an item Identifier where the item identifier originates from an RFID IC associated with an item. The item identification system may also include a learning system configured to receive an image-class pair, where the image-class pair comprises an image of the item and the resolved item class, and to use the image-class pair and a training algorithm to learn an association between the item image and the resolved Item class. The item identification system may additionally include a controller configured to receive a first image and if a first item class corresponding to the first image is available, then the controller is configured to send an image-class pair formed from the first image and the first item: class to the learning system for the learning system to train on using the training algorithm. Otherwise, the controller is configured to send the first image to the learning system for the learning system to provide the first item class.

[0009] These and other features and advantages will be apparent from a reading of the following detailed description and a review of the associated drawings. It is to be understood that both the foregoing general description and the following detailed description are explanatory only and are not restrictive of aspects as claimed.BRIEF DESCRIPTION OF DRAWINGS

[0010] The following Detailed Description proceeds with reference to the accompanying drawings, in which:[00.1 1] FIG, l is a block diagram of components of an RFID system.

[0012] FIG. 2 is a diagram showing components of a passive RFID tag, such as a tag that can be used in the system of FIG, 1.

[0013] FIG. 3 is a conceptual diagram for explaining a half-duplex mode of communication between the components of the RFID system of FIG. 1.

[0014] FIG.4 is a block diagram showing a detail of an RFID tag, such as the one shown in FIG. 2.

[0015] FIG, 5A and 5B illustrate signal paths during tag-to-reader and reader~to~tag communications in the block diagram of FIG. 4,

[0016] FIG. 6 is a block diagram showing a detail of an RFID reader system, such as the one shown in FIG. 1.

[0017] FIG. 7 depicts an item processing system, according to examples.

[0018] FIG. 8 is a block diagram showing how RFID-derived information may be used to train, using supervised learning techniques, machine learning models for item identification, according to examples.

[0019] FIG, 9 is a block diagram showing a process for RFID-assisted training of machine learning models for item identification, according to one example,

[0020] FIG. 10 is a flow diagram illustrating the training of an image classifier using a continual learning process, according to examples.

[0021] FIG. 11 illustrates interactions between an RFID reader and an RFID tag, according to examples.

[0022] FIG, 12 is a flow diagram illustrating an example method for RFID -assisted training of an item identification model using a continual learning process, where the item identification model is trained with RFID-derived information from RFID tags on some items, according to examples.

[0023] FIG, 13 Illustrates a block diagram of an example computer program product, according to examples,DETAILED DESCRIPTION OF THE INVENTION

[0024] In the following detailed description, references are made to the accompanying drawings that form a part hereof, and in which are shown by way of illustration specific examples. These examples may be combined, other aspects may be utilized, and structural changes may be made without departing from the spirit or scope of the present disclosure. The following detailed description is therefore not to be taken in a limiting sense, and the scope of the present invention is defined by the appended claims and their equivalents,

[0025] As used herein, "memory" is one of ROM, RAM, SRAM, DRAM, NVM, EEPROM, FLASH, Fuse, MRAM, FRAM, and other similar volatile and nonvolatile information¬ storage technologies. Some portions of memory may be writeable and some not." Instruction" refers to a request to a tag to perform a single explicit action (e,g., write data into memory], " Command" refers to a reader request for one or more tags to perform one or more actions, and includes one or more tag instructions preceded by a command identifier or command code that Identifies the co and and / or the tag instructions. " Program" refers to a request to a tag to perform a set or sequence of instructions (e,g., read a value from memory and, if the read value is less than a threshold then lock a memory word). " Protocol" refers to an industry standard for communications between a reader and a tag (and vice versa], such as the Class- 1 Generation-2 UHF RFID Protocol for Communications at 860 MHz -960 MHz by GS1 EPCgiobal, Inc. (" Gen2 Protocol"), versions 1.2,0, 2.0, and 3,0 of which are hereby incorporated by reference.

[0026] Items are processed by many different entities in different contexts. Some examples include logistics entities placing items into shipping containers based on weight and / or size of the items, merchants processing sales based on items in a transaction, and recycling facilities sorting items based on material composition of the items, amongst many other examples. The processing of an item often begins withidentifying the item or characteristics of the item. In some situations, items and / or their characteristics can be identified manually, such as by workers in a sortation line.However, to increase processing efficiency, items are often identified by automated systems.(0027] One approach to automate item identification is to train a machine learning model such that the trained model can identify an item based on some data about the item. Such data can be obtained by using a sensor of a data capture system to capture data about the item. For example, a camera may capture an image of the item, or a scale may measure the weight of the item. Machine learning models can be trained to identify items by learning associations between item classes and certain characteristics that can be derived from the captured data. An "item ciass" is a category or group of items that share common characteristics, such as physical characteristics (e.g,, color, size, weight, material composition, etc.), manufacturer, area of use (e.g., clothing, automotive, etc.), or any other characteristic that can describe an item or the functions of the item. Item classes may be defined by a user, and may be made as broad or as narrow as desired. If items are highly distinguishable, a user may wish to define item classes which are narrow. Otherwise, a user may wish to define item classes which are broad- In some exampies, item classes may be or may be derived from identifiers such as Global Trade Item Numbers (GTINs), Electronic Product Codes (EPCs), or Stock Keeping Units (SKUs).

[0028] Embodiments supplement machine-iearning-based item identification with information that is obtained from or derived from RFID tags atached to, or embedded in, items. RFID tags are configured to provide information associated with the item they are attached to. Such information may include an item identifier or characteristics of the i tem. In some situations, the information can instead be used to look-up or derive the item; identifier or item: characteristics, The tag-provided information can be used in the training and deployment of machine learning models, such as to validate the outputs of the models.

[0029] FIG. 1 is a diagram of the components of a typical RFID system 100, incorporating examples. An RFID reader 110 and a nearby RFID tag 120 communicate via RF signals 112 and 126, When sending data to tag 120, reader 110 may genefate RF signal 112 by encoding the data, modulating an RF waveform with the encoded data, and transmitting the modulated RF waveform as RF signal 112. In turn, tag 120 may receive. RF signal 112, demodulate encoded data from RF signal 112, and decode the encoded data. Similarly, when sending data to reader 110 tag 120 may generate RF signal 126 by encoding the data, modulating an RF waveform with the encoded data, and causing the modulated RF waveform to be sent as RF signal 126. The data sent between reader 110 and tag 120 may be represented by symbols, also known as RFID symbols, A symbol may be a delimiter, a calibration value, or implemented to represent binary data, such as "0" and "1", if desired. Upon processing by reader 110 and tag 120, symbols may be treated as values, numbers, or any other suitable data representations.

[0030] The RF waveforms transmitted by reader 110 and / or tag 120 may be in a suitable range of frequencies, such as those near 900 MHz, 13.56 MHz, or similar. in some examples, RF signals 112 and / or 126 may include nan-propagating RF Signals, such as reactive near-field signals or similar. RFID tag 120 may be active or battery-assisted (i.e., possessing Its own power source), or passive. In the latter case, RFID tag 120 may harvest power from RF signal 112.

[0031] FiG, 2 is a diagram of an RFID tag 220, which may function as tag 120 of FIG. 1.Tag 220 may be formed on a substantially planar inlay 222, which can be made in any suitable way. Tag 220 includes a circuit which may be implemented as an IC 224. In some examples IC 224 is fabricated in complementary metal-oxide semiconductor (CMOS) technology. In other examples IC 224 may be fabricated in other technologies such as bipolar junction transistor(BIT) technology, metal-semiconductor field-effect transistor (MESFET) technology, and others as will be well known: to those skilled in the art. IC 224 is arranged on inlay 222.

[0032] Tag 220 also includes an antenna for transmitting and / or interacting with RF signals. In some examples the antenna can be etched, deposited, and / or printed metalon Inlay 222; conductive thread formed with or without substrate nonmetallic conductive (such as graphene) patterning on substrate; a first antenna coupled inductively, capacitively, or galvanically to a second antenna; or can be fabricated in myriad other ways that exist for forming antennas to receive RF waves, in some examples the antenna may even be formed in IC 224. Regardless of the antenna type, iC 224 is electrically coupled to the antenna via suitable IC contacts (not shown in FiG. 2). The term "electrically coupled" as used herein may mean a direct electrical connection, or it may mean a connection that includes one or more intervening circuit blocks, elements, or devices. The "electrical" part of the term "electrically coupled" as used in this document shall mean a coupling that is one or more of ohmic / galvanlc, capacitive, and / or inductive. Similarly, the terms"electrically isolated" or "electrically decoupled" as used herein mean that electrical coupling of one or more types (e.g., galvanic, capacitive, and / or inductive) is not present, at least to the extent possible. For example, elements that are electrically isolated from each other are galvanically isolated from each other, capacitively isolated from each other, and / or inductively isolated from each other. Of course, electrically isolated components will generally have some unavoidable stray capacitive or inductive coupling between them, but the Intent of the isolation is to minimize this stray coupling when compared with an electrically coupled path.

[0033] IC 224 is shown with a single antenna port, comprising two IC contacts electrically coupled to two antenna segments 226 and 228 which are shown here forming a dipole. Many other examples are possible using any number of ports, contacts, antennas, and / or antenna segments. Antenna segments 226 and 228 are depicted as separate from IC 224, but in other examples the antenna segments may alternatively be formed on IC 224. Tag antennas according to examples may be designed in any form and are not limited to dipoles. For example, the tag antenna may be a patch, a slot, a loop, a coil, a horn, a spiral, a monopole, microstrip, stripline, or any other suitable antenna.

[0034] Diagram 250 depicts top and side views of tag 252, formed using a strap. Tag 252 differs from tag 220 in that it Includes a substantially planar strap substrate 254 havingstrap contacts 256 and 258. IC 224 is mounted on strap substrate 254 such that the IC contacts on IC 224 electrically couple to strap contacts 256 and 258 via suitable connections (not shown). Strap substrate 254 is then placed on inlay 222 such that strap contacts 256 and 258 electrically couple to antenna segments 226 and 228. Strap Substrate 254 may be affixed to inlay 222 via pressing, an interface layer, one or more adhesives, or any other suitable means.{0035 Diagram 260 depicts a side view of an alternative way to place strap substrate 254 onto inlay 222. Instead of strap substrate 254' s surface, Including strap contacts 256 / 258, facing the surface of inlay 222, strap substrate 254 is placed with its strap contacts 256 / 258 facing away from the surface of inlay 222. Strap contacts 256 / 258 can then be either capacitively coupled to antenna segments 226 / 228 through strap substrate 254, orconductively coupled using a through-via which may be formed by crimping strap contacts 256 / 258 to antenna segments 226 / 228. In some examples,, the positions of strap substrate 254 and inlay 222 may be reversed, with strap substrate 254 mounted beneath inlay 222 and strap contacts 256 / 258 electrically coupled to antenna segments 226 / 228 through inlay 222. Of course, in yet other exampies strap contacts 256 / 258 may electrically couple to antenna segments 226 / 228 through both inlay 222 and strap substrate 254.

[0036] In operation, the antenna couples with RF signals in the environment and propagates the signals to IC 224, which may both harvest power and respond if appropriate, based on the incoming signals and the IC's internal state. If IC 224 uses backscatter modulation, then it may generate a response signal (e.g., signal 126) from an RF signal in the environment (e.g., signal 112) by modulating the antenna's reflectance. Electrically coupling and uncoupling the IC contacts of IC 224 can modulate the antenna's reflectance, as can varying the admittance or impedance of a shunt- connected or series-connected circuit element which is coupled to the IC contacts. If IC 224 is capable of transmitting signals e.g,, has its own power source, is coupled to an external power source, and / or can harvest sufficient power to transmit signals), then IC 224 may respond by transmitting response signal 126, In the examples of FIG. 2,antenna segments 226 and 228 are separate from IC224. In other examples, the antenna segments may alternatively be formed on IC 224.

[0037] An RFID tag such as tag 220 is often attached to or associated with an individual item or the item packaging. An RFID tag may be fabricated and then attached to the item or packaging, may be partly fabricated before attachment to the item or packaging and then completely fabricated upon attachment to the item or packaging, or the manufacturing process of the item or packaging may include the fabrication of the RFID tag. In some examples, the RFID tag may be integrated into the item or packaging. In this case, portions of the item or packaging may serve as tag components. For example, conduct-ve item or packaging portions may serve as tag antenna segments or contacts. Nonconductive item or packaging portions may serve as tag substrates or inlays, If the item^ or packaging Includes integrated circuits or other circuitry, some portion of the circuitry may be configured to operate as part or all of an RFID tag IC. Thus, an ’’RFID! C" need not be distinct from an item, but more generally refers to the item containing an RFID IC and antenna capable of interacting with RF waves and receiving and responding to RFID signals. Because the boundaries between IC, tag, and Item are thus often blurred, the term " RFID IC", " RFID tag'', "tag", "radio IC", ''radio tag IC", or "tag IC" as used herein may refer to the |C, the tag, or even to the item as long as the referenced element is capable of RFID functionality.

[0038] The components of the RFID system of FIG. 1 may communicate with each other in any number of modes. One such mode is called full duplex, where both reader 110 and tag 120 can transmit at the same time. In some examples, RFID system 100 may be capable of full duplex communication. Another such mode, which may be more suitable for passive tags, is called half-duplex, and is described below.

[0039] FIG. 3 is a conceptual diagram 300 for explaining half-duplex communications between the components of the RFID system of FIG. 1, in this case with tag 120 implemented as a passive tag. The explanation is made with reference to a TIME axis, and also to a human metaphor of "talking" and "listening". The actual technical implementations for "talking" and "listening" are now described.

[0040] in a half-duplex communication mode, RFID reader 110 and RFID tag 120 talk and listen to each other by taking turns. As seen on axis TIME, reader 110 talks to tag 120 during intervals designated " R-> T", and tag 120 talks to reader 110 during intervals designated " T-> R". For example, a sample R-> T interval occurs during time interval 312, during which reader 110 talks (block 332) and tag 120 listens (block 342). A following sample T~> R interval occurs during time interval 326, during which reader 110 listens (block 336) and tag 120 talks (block 346). Interval 312 may be of a different duration than interval 326 ~ here the durations are shown approximately equal only for purposes of illustration.

[0041] During interval 312, reader 110 transmits a signal such as signal 112 described in FIG. 1 (block 352), while tag 120 receives the reader signal (block 362), processes the reader signal to extract data, and harvests power from the reader signal. While receiving the reader signal, tag 120 does not backscatter (block 372), and therefore reader 110 does not receive a signal from tag 120 (block 382),

[0042] During interval 326, also known as a backscatter time interval or backscatter interval, reader 110 does not transmit a data-bearing signal, instead, reader 110 transmits a continuous wave (CW) signal, which is a carrier that generally does not encode information. The CW signal provides energy for tag 120 to harvest as well as a waveform that tag 120 can modulate to form a backscatter response signal. Accordingly, during interval 326 tag 120 is not receiving a signal with encoded information (block 366) and: instead modulates the CW signal (block 376) to generate a backscatter signal such as signal 126 described in FIG. 2. Tag 120 may modulate the CW signal to generate a backscatter signal by adjusting its antenna reflectance, as described above. Reader 110 then receives and processes the backscatter signal (block 386).

[0043] FIG, 4 is a block diagram showing a detail of an RFID IC, such as IC 224 in FIG. 2.Electrical circuit 424 may be implemented in an IC, such as IC 224. Circuit 424 implements at least two IC contacts 432 and 433, suitable for coupling to antenna segments such as antenna segments 226 / 228 in FIG, 2, When two IC contacts form the signal input from and signal return to an antenna they are often referred-to as anantenna port. IC contacts 432 and 433 may be made in any suitable way, such as from electrically conductive pads, bumps, or similar, in some examples circuit 424 implements more than two IC contacts, especially when configured with multiple antenna ports and / or to couple to multiple antennas,

[0044] Circuit 424 includes signal-routing section 435 which may include signal wiring, signal-routing buses, receive / transmit switches, and similar that can route signals between the components of circuit 424. IC contacts 432 / 433 may couple galvanically, capacitively, and / or inductively to signal-routing section 435. For example, optional capacitors 436 and / or 438 may capacitively couple IC contacts 432 / 433 to signal-routing section 435, thereby galvanically decoupling IC contacts 432 / 433 from signal-routing section 435 and other components of circuit 424.

[0045] Capacitive coupling (and the resultant galvanic decoupling) between IC contacts 432 and / or 433 and components of circuit 424 is desirable In certain situations. For example, in some RFID tag examples, IC contacts 432 and 433 may galvanically connect to terminals of a tuning loop on the tag. In these examples, galvanically decoupling IC contact 432 from It contact 433 may prevent the formation of a DC short circuit between the IC contacts through the tuning loop.

[0046] Capacitors 436 / 438 may be implemented within circuit 424 and / or partly or completely external to circuit 424. For example, a dielectric or insulating layer on the surface of the IC containing circuit 424 may serve as the dielectric in capacitor 436 and / or capacitor 438. As another example, a dielectric or insulating layer on the surface of a tag substrate (e.g,, inlay 222 or strap substrate 254) may serve as the dielectric in capacitors 436 / 438. Metallic or conductive layers positioned on both sides of the dielectric layer (i.e., between the dielectric layer and the IC and between the dielectric layer and the tag substrate) may then serve as terminals of the capacitors 436 / 438. The conductive layers may include IC contacts (e.g., IC contacts 432 / 433), antenna segments (e.g,, antenna segments 226 / 228), or any other suitable conductive layers.

[0047] Circuit 424 includes a rectifier and PMU (Power Management Unit) 441 that harvests energy from the RF signal incident on antenna segments 226 / 228 to power thecircuits of IC 424 during either or both reade to-tag (R-> T) and tag-to-read er (T-> R) intervals. Rectifier and PMU 441 may be implemented in any way known in the art and may include one or more components configured to convert an alternating-current (AC) or time-varying signal into a direct-current (DC) or substantially time-invariant signal.

[0048] Circuit 424 also includes a demodulator 442, a processing block 444, a memory 450, and a modulator 446, Demodulator 442 demodulates the RF signal received via IC contacts 432 / 433, and may be implemented in any suitable way, for example using a slicer, an amplifier, and other similar components. Processing block 444 receives the output from demodulator 442, performs operations such as command decoding, memory interfacing, and other related operations, and may generate an output signal for transmission. Processing block 444 may be implemented in any suitable way, for example by combinations of one or more of a processor, memory, decoder, encoder, and other similar components. Memory 450 stores data 452 and may be at least partly implemented as permanent or semi-permanent memory such as nonvolatile memory (NVM), EEPROM, ROM, or other memory types configured to retain data 452 even when circuit 424 does not have power. Processing block 444 may be configured to read data from and / or write data to memory 450.

[0049] Modulator 446 generates a modulated signal from the output signal generated by processing block 444, In one embodiment, modulator 446 generates the modulated signal by driving the load presented by antenna segment(s) coupled to IC contacts 432 / 433 to form a backscatter signal as described above, In another embodiment, modulator 446 includes and / or uses a transmitter to generate and transmit the modulated signal via antenna segment(s) coupled to IC contacts 432 / 433, Modulator 446 maybe implemented in any suitable way, for example using a switch, driver, amplifier, and other similar components. Demodulator 442 and modulator 446 may be separate components, combined in a single transceiver circuit, and / or part of processing block 444,

[0050] In some examples, particularly In those with more than one antenna port, circuit 424 may contain multiple demodulators, rectifiers, PM Us, modulators, processing blocks, and / or memories.

[0051] FIG. 5A shows version 524-A of components of circuit 424 of FIG. 4, further modified to emphasize a signal operation during a R-> T interval (e.g., time interval 312 of FIG, 3), During the R-> T Interval, demodulator 442 demodulates an RF signal received from IC contacts 432 / 433. The demodulated signal is provided to processing block 444 as CJN, which in some examples may include a received stream of symbols. Rectifier and PMU 441 may be active, for example harvesting power from an incident RF waveform and providing power to demodulator 442, processing block 444, and other circuit components. During the R-> T interval, modulator 446 is not actively modulating a signal, and in fact may be decoupled from the RF signal. For example, signal routing section 435 may be configured to decouple modulator 446 from the RF signal, or an impedance of modulator 446 may be adjusted to decouple it from the RF signal.

[0052] FIG. 5B shows version 524-B of components of circuit 424 of FIG. 4, further modified to emphasize a signal operation during a ~ > R interval (e.g., time interval 326 of FIG. 3). During the T-> R interval, processing block 444 outputs a signal C„OUT, which may include a stream of symbols for transmission. Modulator 446 then generates a modulated signal from C_OUT and sends the modulated signal via antenna segment(s) coupled to IC contacts 432 / 433, as described above. During the T-> R interval rectifier and PMU 441 may be active, while demodulator 442 may not be actively demodulating a signal. In some examples., demodulator 442 may be decoupled from the RF signal during the T-> R interval. For example, signal routing section 435 may be configured to decouple demodulator 442 from the RF signal, or an impedance of demodulator 442 may be adjusted to decouple it from the RF signal.

[0053] In typical examples, demodulator 442 and modulator 446 are operable to demodulate and modulate signals according to a protocol, such as the Gen2 Protocol mentioned above. In examples where circuit 424 includes multiple demodulators modulators, and / or processing blocks, each may be configured to support differentprotocols or different sets of protocols. A protocol specifies, in part, symbol encodings, and may include a set of modulations, rates, timings, or any other parameter associated with data communications. A protocol can be a variant of an internationally ratified protocol such as the Gen2 Protocol, for example including fewer or additional commands than the ratified protocol calls for, and so on. In some instances, additional commands may sometimes be called custom commands.

[0054] FIG, 6 depicts an RFID reader system 600 according to examples. Reader system 600 is configured to communicate with RFID tags and optionally to communicate with entitles external to reader system 600, such as a service 632. Reader system 600 includes at least, one reader module 602, configured to transmit signals to and receive signals from RFID tags. Reader system 600 further includes at least one local controller 612, and in some examples includes at least one remote controller 622. Controllers 612 and / or 622 are configured to control the operation of reader module 602, process data received from RFID tags communicating through reader module 602, communicate with external entities such as service 632, and otherwise control the operation of reader system 600.

[0055] In some examples, reader system 600 may include multiple reader modules, local controllers, and / or remote controllers. For example, reader system 600 may include at least one other reader module 610, at least one other local controller 620, and / or at least one other remote controller 630. A single reader module may communicate with multiple local and / or remote controllers, a single local controller may communicate with multiple reader modules and / or remote controllers, and a single remote controller may communicate with multiple reader modules and / or local controllers. Similarly, reader system 600 may be configured to communicate with multiple external entities, such as other reader systems (not depicted] and multiple services (for example, services 632 and 640),[0056[ Reader module 602 includes a modulator / encoder block 604, a demodulator / decoder block 606, and an interface block 608, Modulator / encoder block 604 may encode and modulate data for transmission to RFID tags, Demodulator / decoder block606 may demodulate and decode signals received from RFID tags to recover data sent from the tags. The modulation, encoding, demodulation, and decoding may be performed according to a protocol or specification, such as the Gen2 Protocol. Reader module 602 may use interface block 608 to communicate with local controller 612 and / or remote controller 622, for example to exchange tag data, receive Instructions or commands, or to exchange other relevant information.

[0057] Reader module 602 and blocks 604 / 606 are coupled to one or more antennas and / or antenna drivers (not depicted),, for transmitting and receiving RF signals, In some examples, reader module 602 is coupled to multiple antennas and / or antenna drivers. In these examples, reader module 602 may transmit and / or receive RF signals on the different antennas in any suitable scheme. For example, reader module 602 may switch between different antennas to transmit and receive RF signals, transmit on one antenna but receive on another antenna, or transmit and / or receive on multiple antennas simultaneously, in some examples, reader module 602 may be coupled to one or more phased-array or synthesized-beam antennas whose beams can be generated and / or steered, for example by reader module 602, local controller 612, and / or remote controller 622.

[0058] Modulator / encoder block 604 and / or demodulator / decoder block 606 may be configured to perform conversion between analog and digital signals. For example, modulator encoder block 604 may convert a digital signal received via interface block 608 to an analog signal for subsequent transmission, and demodulator / decoder block 606 may convert a received analog signal to a digital signal for transmission via interface block 608.[00591 local controller 612 includes a processor block 614, a memory 616, and an interface 618. Remote controller 622 includes a processor block 624, a memory 626, and an interface 628. Local controller 612 differs from remote controller 622 in that local controller 612 is collocated or at least physically near reader module 602, whereas remote controller 622 is not physically near reader module 602.

[0060] Processor blocks 614 and / or 624 may be configured to, alone or in combination, provide different functions. Such functions may include the control of other components, such as memory, interface blocks, reader modules, and similar; communication with other components such as reader module 610, other reader systems, services 632 / 640, and similar; data-processing or algorithmic processing such as encryption, decryption, authentication, and similar; or any other suitable function. In some examples, processor blocks 614 / 624 may be configured to convert analog signals to digital signals or vice-versa, as described above in relation to blocks 604 / 606processor blocks 614 / 624 may also be configured to perform any suitable analog signal processing or digital signal processing, such as filtering, carrier cancellation, noise determination, and similar.

[0061] Processor blocks 614 / 624 may be configured to provide functions by execution of instructions or applications, which may be retrieved from memory (for example, memory 616 and / or 626) or received from some other entity. Processor blocks 614 / 624 may be implemented in any suitable way. For example, processor blocks 614 / 624, or indeed any processors or process blocks described herein, may be implemented using digital and / or analog processors such as microprocessors and digital-signal processors (DSPs); controllers such as microcontrollers; software running in a machine such as a general purpose computer; programmable circuits such as field programmable gate arrays (FPGAs), field-programmable analog arrays (FPAAs), programmable logic devices (PLDs), application specific integrated circuits (ASIC), any combination of one or more of these; and equivalents. The processor or processor block may include one or more levels of caching, such as a level cache memory, a processor core, and one or more registers. In some examples, the processor or processor block may include an arithmetic logic unit (ALU), a floating point unit (FPU), a digital signal processing core (DSP Core), or any combination thereof. An example memory controller may also be used with or incorporated in the processor or processor block.

[0062] Memories 616 / 626 are configured to store information, and may be implemented in any suitable way,, such as the memory types described above, anycombination thereof, or any other known memory or information storage technology. Memories 616 / 626 may be implemented as part of their associated processor blocks (e,g,, processor blocks 614 / 624) or separately. Memories 616 / 626 may store instructions, programs, or applications for processor blocks 614 / 624 to execute.Memories 616 / 626 may also store other data, such as files, media, component configurations or settings, etc.

[0063] In some examples, memories 616 / 626 store tag data. Tag data may be data read from tags, data to be written to tags, and / or data associated with tags or tagged items. Tag data may include identifiers for tags such as electronic product codes (EPCs), tag identifiers (TIDs), or any other information suitable for identifying individual tags. Tag data may also include tag passwords, tag profiles, tag cryptographic keys (secret or public), tag key generation algorithms, and any other suitable information about tags or items associated with tags.

[0064] Memories 616 / 626 may also store information about how reader system 600 is to operate. For example, memories 616 / 626 may store information about algorithms for encoding commands for tags, algorithms for decoding signals from tags, communication arid antenna operating modes, encryption / authentication algorithms, tag location and tracking algorithms, cryptographic keys and key pairs (such as public / private key pairs) associated with reader system 600 and / or other entities, electronic signatures, and similar,

[0065] Interface blocks 608, 618, and 628 are configured to communicate with each other and with other suitably configured interfaces. The communications between interface blocks occur via the exchange of signals containing data, instructions, commands, or any other suitable information. For example, interface block 608 may receive data to be written to tags, information about the operation of reader module 602 and its constituent components, and similar; and may send data read from tags, interface blocks 618 and 628 may send and receive tag data, information about the operation of other components, other information for enabling local controller 612 and remote controller 622 to operate in conjunction, and similar. Interface blocks608 / 618 / 628 may also communicate with external entities, such as services 632, 640, other services, and / or other reader systems.

[0066] interface blocks 608 / 618 / 628 may communicate using any suitable wired or wireless means. For example, interface blocks 608 / 618 / 628 may communicate over circuit traces or interconnects, or other physical wires or cables, and / or using any suitable wireless signal propagation technique. In some examples, interface blocks 608 / 618 / 628 may communicate via an electronic communications network, such as a local area network (LAN), a metropolitan area network (MAN), a wide area network (WAN), a network of networks such as the internet. Communications from interface blocks 608 / 618 / 628 may be secured, for example via encryption and other electronic means, or may be unsecured,

[0067] Reader system 600 may be implemented in any suitable way. One or more of the components in reader system 600 may be implemented as integrated circuits using CMOS technology, BIT technology, MESFET technology, and / or any other suitable physical implementation technology. Components may also be implemented as software executing on general-purpose or application-specific hardware.(0068] In one embodiment, a "reader" as used In this disclosure may include at least one reader module like reader module 602 and at least one local controller such as local controller 612. Such a reader may or may not include any remote controllers such as remote controller 622, A reader including a reader module and a local controller may be implemented as a standalone device or as a component in another device. In some examples, a reader may be implemented as a mobile device, such as a handheld reader, or as a component in a mobile device such as a laptop, tablet, smartphone, wearable device, or any other suitable mobile device.

[0069] Remote controller 622, if not included In a reader, may be implemented separately. For example, remote controller 622 may be implemented as a local host, a remote server, or a database, coupled to one or more readers via one or more communications networks. In some examples, remote controller 622 may be implemented as an application executing on a cloud or at. a datacenter.

[0070] functionality within reader system 600 may be distributed in any suitable way. For example, the encoding and / or decoding functionalities of blocks 604 and 606 may be performed by processor blocks 614 and / or 624. in some examples, processor blocks 614 and 624 may cooperate to execute an application or perform some functionality, One of local controller 612 and remote controller 622 may not implement memory, with the other controller providing memory,

[0071] Reader system 600 may communicate with at least one service 632, Service 632 provides one or more features, functions, and / or capabilities associated with one or more entities, such as reader systems, tags, tagged items, and similar. Such features, functions, and / or capabilities may include the provision of information associated with the entity, such as warranty information, repair / replacement information, upgrade / update information, and similar; and the provision of services associated with the entity, such as storage and / or access of entity-related data, location tracking for the entity, entity security services (e.g., au thentication of the entity), entity privacy services (e.g,, who is allowed access to what information about the entity), and similar. Service 632 may be separate from reader system 600, and the two may communicate via one or more networks.

[0072] In some examples, an RFID reader or reader system implements the functions and features described above at least partly in the form of firmware, software, or a combination, such as hardware or device drivers, an operating system, applications, and the like. In some examples, interfaces to the various firmware and / or software components may be provided. Such interfaces may include application programming interfaces (APIs), libraries, user interfaces (graphical and otherwise), or any other suitable interface. The firmware, software, and / or interfaces may be implemented via one or more processor blocks, such as processor blocks 614 / 624. In some examples, at least some of the reader or reader system functions and features can be provided as a service, for example, via service 632 or service 640.

[0073] The various systems and interfaces described herein may be implemented by a computing device, A computing device as mentioned herein may be any type ofcomputer and include a memory, a communication subsystem, and one or more processors or processor blocks, among other components. For exampie,, the computing device may be a server such as an on-premise server, a cloud server, or similar.Examples of computing devices are not limited to servers. Any type of computing device, desktop, laptop, mobile, vehicle-mount, body-worn, etc,, may be used to implement or communicate with an item information system, in some cases, the computing device may be distributed, that is, multiple computing devices in communication with each other may coordinate execution of or communication with the item identification system,

[0074] Depending on configuration, the memory may be of any type including but not limited to volatile memory (such as RAM), non-volatile memory (such as ROM, flash memory, etc.) or any combination thereof, as described above. The memory may store data associated with an operating system, one or more applications, and program data,

[0075] The communication subsystem of the computing device may facilitate data exchange with other computing devices, input devices, and output devices, such as an item identification system. Data exchange may be facilitated using wired and / or wireless communication, standardized or proprietary communication protocols. Data exchange may take place over a communications medium, such as, but not limited to, a digital and / or an analog communication medium (e.g., a fiber optic cable, a waveguide, a wired communications link, a wireless communication link, etc.). The communication subsystem may be similar to the interface blocks described above in reference to FIG, 6 and / or the interfaces described below,

[0076] In general, an RFID tag IC has or stores one or more identifiers, which as described above is a number or bit sequence that identifies the tag IC or associated i tem. The identifier may contain information about the tag IC or item (e.g,. a TID or an EPC) or may be used to look up information about the tag IC or item. The tag IC identifier may uniquely identify (at least within the constraints of a finite-length identifier) or be used to uniquely identify the tag IC or associated item.

[0077] FIG.7 depicts an item processing system 700, according to examples. The item processing system 700, which may include or be implemented in a facility 710 with a conveyor system 712, includes a data capture system 714 implementing one or more sensors, an RFID reader system 716, and Items 718 to be processed, where at least some of the items 718 have associated RFID tags 720 that may be attached to or incorporated in the items 718. The data capture system 714 and the RFID reader system 16 may be communicatively coupled to respective servers 702 and 704. Servers 702 / 704 may be located at the facility 710 or may be remote. The item processing system 700 may be configured to use machine learning to identify and process items 718 based on relevant characteristics. ''Characteristics" are features of items that can be either meas ed or observed, and can be used to identify items, or the manner in which items should be processed. In some examples., the item processing system 700 Is configured to first identify the items 718 by identifying item classes of the items 718 ~ fpr example, based on characteristics of the items 718 measu red using the capture system 714 - and then to determine how to process the items 718 based on their identified item classes. For example, the item processing system 700 may first identify the item class of the items 718, The item processing system 700 may then use the conveyor system 712 to sort the items 718, based on identified item classes, into different containers and / or to different processing streams. For example, the item processing system 700 may cause the conveyor system 712 to direct items having a first item class to a first location and items having a second item class to a second location.

[0078] The item processing system 700 may be configured to identify items 718 based on data captured by sensors of the data ca ture syste 714, information read by the RFID reader system 716, data received from external devices, or any combination of the foregoing. The data capture system 714 may use sensors (e.g., cameras, infrared scanners, weight scales, microphones, etc.) to capture different types of data about the items. For example, the data capture system 714 may use a camera and a scale to capture an image and the weight of items 718.

[6079] in some exampies, the item processing system 700 may indude one or more controllers, processors, processing blocks, and / or servers configured to (a) receive information from the data capture system 714, the RFID reader system 716, servers 702 / 704, and / or other associated externa! devices, (b) determine the identities of items 718 and how the items 718 should be processed, and / or (c) control the conveyor system 712 to process the items 718 accordingly. In some examples, the item processing system 700 may implement the controller entirely on, or distribute the controller between, the server 702, the server 704;and / or another server or device in communication with the server 702 and the server 704.

[0080] The server 702 may implement an item identification model that is configured to use data about the items 718 captured by sensors of the data capture system 714 to identify the items 718. The server 702 may represent one or more servers executing the item identification model, which may include machine learning, data and image processing, sorting contra!, and other subsystems. Each of the subsystems may include software and / or hardware components such as processors, controllers, communication systems, etc. The item identification mode! uses machine learning techniques to determine information related to item identification and / or to processing of items, based on data provided to the item identification model - for example, by learning associations between item classes and data captured by sensors of the data capture system 714. Such information may include an item class, physical characteristics, or processing steps associated with the item. Once the item identification model has determined information relating to.classification of an item, it can notify the item processing system 700 (for example, by providing the determined information via the server 702], which can then seek additions! information about the item and / or process the item accordingly. For example, in the context of processing items at a point of sale (POS) terminal, the item identification model may provide an estimation of the identity of items that a customer wishes to purchase to the item processing system 760 which can then determine how the transaction should proceed. In one such example, the item identification model may determine that some data input to the model (e.g., an image)depicts items including a restricted good, such as alcohol, and a frozen food 'item, such as ice cream. The item identification model can provide the determination to the item processing system 700, which then determines that the restricted good requires the customer to verify their age in order for the transaction to proceed, and that the frozen food item should be packaged in an insulated container after the transaction is complete.

[0081] In some situations, data captured by sensors of the data capture system 714 can be supplemented with information retrieved from RFID tags (" RFID tag information”) to assist the item identification model with the identification of items and / or to train machine learning models used by the item identification model. For example, the RFID reader system 716 may retrieve RFID tag information from the tags 720 as described above. In some examples, the RFID tag information may uniquely identify the associated item, for example, by including the entirety ora portion of a tag or item identifier. In other examples, the RFID tag information may broadly identify the associated item, such as by including an item class or physical characteristics of an item and may also include processing steps associated with the item.

[0082] In some situations, RFID tag information may not include information about the identity or processing of an item, but can instead be used to look up relevant or additional information about the item, for example, via an external entity such as a resolver system or service. This entity is depicted in FIG. 7 by server 704. In one example, the RFID tag information may include an identifier for a tag or item that does not in and of itself provide any information about the tag or item. However, when the identifier is provided to a resolver system or service, the resolver system or service may then use the identifier to look up or resolve information specific to the tag or item. RFID tag information and any information resolved from the RFID tag information may individually or collectively be referred to herein as " RFID-derived information".

[0083] When available, RFID-derived information can be used to confirm: and / or refute determinations made by the item identification model of the server 702. If the RFID- derived information of an item refutes (i.e., does not match or correspond to) thedetermination made by the item identification model, then the item processing system 700 can be configured to alert an external device of the mismatch. The item processing system 700 can also be configured to determine whether to abide by the RFID-derived information or the determination and proceed accordingly; The alert may include the RFID-derived information and the determination. In other examples, if the item processing system 700 finds that the RFID-derived information of the item refutes the determination made by the item identification model, it may route the item to a different processing stream (e.g., for manual identification and processing), in some situations, RFID-derived information may be considered to be more accurate than determinations made by the item identification mode! and if so, then the item processing system 700 may give precedence to the RFID-derived information. This may be the case, for example, because the item identification model may be unable to obtain a "good” determination of the identity or processing of an item if it is misshapen or damaged and the item identification model is only trained based on data about well- built or non-damaged items, or if one or more sensors of the data capture system 714 are malfunctioning or not calibrated well.

[0084] When both captured data and RFID-derived information of an item are available for a given item, then the captured data and RFiD-derived information can be used to train the item: identification model. For example, if a determination based on some captured data about an item made by the item identification model matches the RFID- derived information, then the captured data and either, or both, the determination or the RFID-derived information can be used to reinforce the machine learning aspects of the item identification model. On the other hand, if the determination does not match the RFID-derived information, then the captured data and the RFID-derived information can be used to train the machine learning aspects of the item identification model to improve future item class determinations, assuming that the RFID-derived Information is correct and takes precedence.

[0085] FIG. 8 is a block diagram 800 showing how RFID-derived information may be used to train, using supervised learning techniques, machine learning models for item identification, according to examples.

[0086] Diagram 800 depicts an item identification server 802 implementing a machine learning model configured to identify items by learning associations between item classes and captured data. The item identification server 802 is configured to receive data about items captured by a data capture system 814. The item identification server 802 includes or implements a learning module 822, In hardware or software form, configured to identify items from input data. The learning module 822 may use one or more machine learning techniques, such as those used in image classifiers, image segmentors, or recommender systems., all described below, to perform determinations of item identities.

[0087] The machine learning portions of the learning module 822 are trained by a training module 824 in a continual learning process, which may be implemented by the item i entification server 802 or by another entity. A continual learning process is a training process in which a machine learning model Is trained to incrementally learn how to identify items, either previously known or unknown to the model, while maintaining its prior learnings on how to identify items, The training module 824 may continue to train the learning module 822 until it is able to reliably identify items. The reliability of the learning module 822 can be measured by one or more metrics. Such metrics can include an accuracy and a precision, which are measures of the correctness of identifications made by the learning module 822, and a recall, which is a measure of the quantity of identifications made by learning module 822. For example, it may be satisfactory that the item identification model provides an item class with a 68%, 70%, 80%, 90%, 95% accuracy, precision, or recall. The training module 824 is configured to train the machine learning portions of the learning module 822 using one or more data sets. In supervised learning examples, the training module 824 may use a training data set including data about various items and their respective item class. In some examples, training data may be obtained by, for example,, the item processing system700 -;.e., the data captured by sensors of the data capture system 714 and item class from RFID-derived information retrieved by the RFID reader system 716. in some examples, the training module 824 may be configured to train the learning module 322 offline, meaning that the training occurs while the learning module 822 is not available to perform identifications of items. In other examples, the training module 824 may be configured to train the learning module 822 online, meaning that the training occurs in real-time, or while the learning module 822 is available to perform identifications of items.

[0088] Diagram 800 also depicts an RFID server 804 configured to determine RFID- derived information 826. The RFID server 804 is coupled to an RFID reader system 816 that is configured to read or retrieve RFID tag information from the RFID tags associated with items. In diagram 800, RFID reader system 816 retrieves RFID tag information, described above, from an RFID tag 820 associated with an item 818. In some situations,, the RFID tag information from the RFID tag 820 may already include relevant information (e.g., identifiers, item class, processing information, etc.). In other situations, the RFID server 804 may use the RFID tag information to obtain RFiD-derived information 826 in a process similar to that described above (e.g.fresolving RFID tag information itself, or by communicating with some other resolver system or service). Communications between the RFID reader system 816, the RFID server 804, and / or any external server may be secured, for example using encryption or digital signatures, or in plaintext or otherwise unsecured.

[0089] in any event, the RFID-derived information 826 can be used in several ways, as described above with reference to FIG.7. For example, if the RFiD-derived information 826 refutes a determination made by the learning module 822, it may be used in an alert. In another exampie, the training module 824 may use the RFID-derived information 826, along with the data about the item 818 captured by sensors of the data capture system 814, to reinforce or otherwise further train the machine learning portions of the learning module 822.

[0090] FIG, 9 is a block diagram 90Q showing a process for RFID-assisted training of machine learning models for item identification, according to one exampie. Diagram 900 in FIG. 9 shows how an initial model 902 may undergo three stages - training 904, evaluation 906, and fine tuning 908 - to generate or arrive at a trained model 910. The trained model 910 may be used in the identification of items to be processed at a facility, and may also be periodically or continuously re-trained to increase accuracy. The initial model 902 may be any suitable machine learning item identification model, such as an item classifier, an image segmentor, or a recommender system. Training 904 may include the use of training data 914 including data-class pairs of items that were previously successfully identified. The data in the data-class pairs may be captured by a data capture system of the facility and the associated item class may be obtained from ftFlD^derived information. Evaluation 906 may include testing of the initially trained model using test data 916. The test data 916 may include data-class pairs obtained at the facility. Fine tuning 908 may include adjustment of the tested model using validation data 918. The validation data 918 may include data-class pairs of previously successfully identified items. For example, after the initial model 902 goes through evaluation 906, it may be in live-use and validation data 918 from successful examples of data-class pairs from item identifications may be accumulated. The validation data 918 may be used for fine tuning 908 when the model is taken offline or while it is online.

[0091] Machine learning algorithms, which are part of artificial intelligence (Al), build a model based on sample data, or training data, in order to make predictions or decisions without being explicitly programmed to do so. Machine learning methods are commonly divided into three broad categories, depending on the type of feedback received by the learning system: supervised learning, unsupervised learning, and reinforcement learning. Supervised learning uses example inputs and corresponding desired outputs to generate a general rule that maps inputs to outputs. Unsupervised learning relies on detecting patterns in data without exampie input-output pairs. Reinforcement Seaming provides feedback to a machine learning algorithm as It navigates the data set(s).Logistic Regression and Neurai Network Back Propagation are examples of techniquesused in supervised learning, which typically uses iterative optimization of an objective function, Aprior I algorithms and K-Means are examples of techniques used in unsupervised learning, which may utilize clustering, dimensionality reduction, and association rule learning. There is also a mixture of supervised and unsupervised learning (semi-supervised learning) algorithms, where the input data is a mixture of labeled and unlabeled examples. Reinforcement learning algorithms may start with a Markov decision process,

[0092] A model in machine learning, which is trained using training data and then is used to process additional data, may also vary depending on the algorithm used to build the model and / or type of data being processed. Example models include artificial neural networks, decision trees, regression models, and Bayesian networks. Deep learning algorithms are a recent development related to artificial neural networks. Deep learning algorithms build models with larger and more complex neural networks and are used for image or video processing. Popular deep learning algorithms include Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), Long Short-Term Memory Networks (LSTMs), Stacked Auto-Encoders, Deep Boltzmann Machine (DBM), and Deep Bel ief N et rks ( DS ).

[0093] As mentioned above, the selected model may be trained with the selected machine learning algorithm and be fine-tuned with the validation data 918 derived from items that were previously successfully identified.

[0094] in some examples, training or fine-tuning the item identification model includes training the item identification model to identify particular classes of items based on some data along with RFID-derived information originating from RFID tags associated with the Items, which includes an Item class or has been resolved into an item class. If the same item or the same type of item is encountered at a later time, but without an RFID tag [e,g., the RFID tag may be broken, removed, or not be included in a different but similar item), the trained model 910 may still be able to identify the item because it has already been trained to recognize associations between certain data and a corresponding item class. Similarly, if an item is misshapen or damaged but retains itsRFID tag, then even if the trained model 910 cannot identify the item, it can be processed based on RFtD-derived information, in this situation, the data captured of the misshapen / damaged item and the RFID-derived information can then be used to further train the item identification model., such that it may be able to identify even misshapen or damaged items.

[0095] FlG. 10 is a flow diagram 1000 illustrating the training of an image classifier using a continual learning process, according to examples. The steps of FIG, 10 are described with reference to the components of the item processing system 700 of FIG. 7, with the presumption that the item processing system 700 is configured to identify items by using a "vision-based" (i.e., uses images) item identification model In particular, an image classifier™ a type of machine learning model that takes an image as input to the model and identifies a class of an object present in the image - can be implemented by the server 702. Image classifiers can be configured to perform different types of classifications, such as binary classifications (e.g., this is / is not an image of a dog), multictass classification (e,g., this is an image of a dog / cat / elephant / etc.), multilabel classification (e.g., this is an image containing a dog, cat, and an elephant), and hierarchical classifications (e.g., this is a n image of a mammal and this is an image of a dog). The image classifier may be implemented by a neural network, such as in a Convolutional Neural Network scheme, that is defined by a trainable set of weights and biases. The image classifier may have various performance metrics by which its quality is determined. Such metrics may include an accuracy, a precision, and a recall as described above.

[0096] An interface or controller (not shown in FIG. 7) in communication with at least the server 702 and the server 704 can be configured to cause the steps described below to be performed.

[0097] At block 1002., the data capture system 714 may capture an image of an item. Of Course, the data capture system 714 may include additional sensors configured to capture additional data about the item 718 (e.g., a scale to capture weight, a different, multispectral camera to capture image data at different wavelengths, etc.).

[0098] At block 1004, the server 702 may run inference on the Innage using the image classifier (Le use the image as input to the image classifier) to obtain an item classification result, in embodiments, the image classifier can be configured such that it is able to identify items of multipie different classes (configured to perform multiclass, multilabel, or hierarchical classifications) or the image classifier could include multiple sub-classtfiers with eachsub-classifier configured to identify one item class.

[0099] At block 1006, the RFID reader system 716 may determine if RFID tag information of the item is available. For example, the RFID reader system 716 may perform the determination by initiating an inventorying process with RFID tags 720 within its range. In some examples, the Item processing system 700 may be configured to limit the range that the RFID reader 716 searches for RFID tags to those on the conveyor system 712, such that other information of other RFID tags not of interest is not captured (I. e., such that only RFID tags 720 are interrogated).

[0100] At block 1008, if RFID information is available, the controller or the RFID reader system 716 may obtain RFID-derived information. For example, the controller or reader system 716 may use RFID tag information to obtain RFID-derived information, as described above, which includes an item class of the item 718 associated with the tag 720.

[0101] At block 1010, the server 702 may update a data set to include the captured image and RFiD-clerived information. The image and an item class from the RFID-derived information may be stored in the data set as a "image-class" pair. The data set may be used as a training, test, or validation data set In some exampies, the data set may also include image-class pairs of previously Identified Items.

[0102] At block 1012, the server 702 may determine if enough updates to the data set are available. In some examples, the server 702 may be configured to cause the image classifier to be trained on each update of the data set. in other examples, the server 702 may be configured to cause the image classifier to be trained after a sufficient number of updates to the data set have been made or after a sufficient percentage increase in the size of the data set. In further examples, the server 702 may be configured to causethe image classifier to be trained after some time period has elapsed (e.g., after one hour, one day, one week, etc.).

[0103] At block 1014, in response to determining that enough updates to the data set are available, the server 702 may train the image classifier using a partition or portion of the data set to obtain an updated image classifier. The training of the image classifier may refer to training or retraining of the image classifier. In some examples, the server 702 may train the image classifier using a partition that only includes image-class pairs that have not been previously used for raining. The training of the image classifier may result in an updated set of values for trainable weights and biases for the image classifier. For example, the structure of a Convolutional Neural Network (CNN) may remain the same prior to the training and after the training, but the specific values of weights and biases at certain locations of the CNN may differ.

[0104] At block 1016, the server 702 may test the updated image classifier using a second partition or portion of the data set. For example, the server 702 may use a partition including only the previously obtained image-class pairs of the data set (e.g., the data set before the update) to measure various metrics of the u pdated image classifier, such as the accuracy, precision, and / or recall of the updated image classifier.

[0105] At block 1018, the server 702 may determine If the updated image classifier performs better than the original image classifier. For example, the server 702 may determine metrics of the updated image classifier and compare the metrics of the updated: image classifier to the metrics of the original image classifier (which may have been measured and stored previously).

[0106] In response to determining that the updated image classifier performs better than the original image classifier, the server 702 may proceed with the updated image classifier. The item processing system 700 may thereafter identify and process items, with the server 702 using the updated image classifier. However, if the updated image classifier does not perform better than the original classifier, the Item processing system 700 may revert to the original image classifier.

[0107] The process described above illustrates the training of an item classifier based on images and RFID-derived information of items. Broadly, the process includes obtaining data (i.e,, images and RFID-derived information) to update a data setter training and / or five-use of a machine learning model training the machine learning model using the updated data set, and evaluating whether the trained machine learning model petdorms better than the machine learning model prior to the training. While the specifics of each part of the process may differ between other implementations using different machine learning models and techniques, the broad process remains largely similar.

[0108] In other examples, the server 702 may implement a machine learning model that uses image-based segmentation, such as a semantic segmentation system. One such example may be an Image segmentor, An image segmentor can identify different regions of a provided images (e.g., individual, or regions of, pixels of images) and assign labels to the identified regions. In another example, the server 702 may implement a machine learning model that learns to numerically represent items as embeddings, such as a recommender system, which learns how to represent item classes as embeddings and identifies one or more item classes that are similar to data it receives as input. Different machine learning models may result in determinations that output a specific item class (e.g., as in image classifie s or i age seg entors) or a set of probabilities of different item classes (e.g., as In recommender systems). Further, some machine learning models may be configured to broadly identify Items, outputting item classes that are broad, such as food, metal, fabric, cotton, clothing, T-shirt, and jeans. Others may be configured to identify more specific types of items, such as blue pants, brown pants, and black pants.

[0109] In further examples, machine learning models may be configured to identify items (additionally) based on other data, such multispectral images, physical properties (e.g<, size, weight, material composition, chemical formula, etc.), retail price, manufacturer, or any other data that can be captured by sensors of a data capture system. RFID-derived information may be used to train such machine learning models and may be used to verify the identity of items during live use of the item processingsystem (e.g., by comparing the item class determined by the machine teaming model to the RFID-derived item class).

[0110] FIG. 11 illustrates interactions between an RFID reader 1102 and an RFID tag 1104, according to examples,

[0111] Diagram 1100 depicts an RFID reader system 1102 wirelessly communicating With an RFID tag 1104 to retrieve RFID tag information from the tag. The RFID reader system 1102 may transmit an inventorying command (e.g,, a Query command according to the Genl Protocol) to the RFID tag 1104 which provides RFID tag information, such as item information and tag information in response. The received RFID tag information may already include information relevant to the identification or the classification of an item associated with the RFID tag 1104 (not shown). In some examples, the RFID reader system 1102 may also estimate a location of the RFID tag 1104, use the RFID tag information and the estimated location to classify the associated item, and or provide the RFID tag information and the estimated location to an item identification server 1108 to be used as validation data in the training of an item identification model, such as an image classifier described above.

[0112] A resolver system, or alternatively a resolver service, may be configured to resolve an item class from RFID tag information and provide the resolved item class as RFID-derived information to a controller or interface that is configured to facilitate item identification. The RFID reader system 1102 or the database 1106 may implement such a resolver system, in the example shown by FIG. 11, the RFID reader system 1102 provides the item information to the item identification server 1108, acting as the cantroller which receives RFI D-derived information. In other examples, a different device may implement the controller.

[0113] RFID tag location can be useful for distinguishing between different items that may be present within a data capture region of a data capture system 1110. In other examples, the RFID reader system 1102 may be constrained to only read RFID tags in specific locations, to avoid reading stray tags outside of the data capture region of the data capture system 1110.

[0114] As described above, the data capture system 1110 may capture data, such as an image, about an item associated with the tag 1104. The item identification server 1108 may itself facilitate the item identification process and may, using a trained item identification model, use the image of the item as input to the model to determine an item class of the item based on prior learning of associations between item ciasses and item images. The item identification server 1108 may compare the determined item class to the resolved item class. The item identification server 1108 may provide the resuit of the comparison to an external computer or may itself use the comparison to perform different processing steps for the item. For example, if the comparison indicates the item class determined by the item identification model and the resolved item: class match, the item identification server 1108 may proceed to process the item according to the item class and provide the image and the item class of the item as training data to the item identification model. In another example, the comparison may indicate that the determined item class and the resolved item class do not match, and the item identification server 1108 may provide an alert of the mismatch to an external device. The aiert may include an indication of the item class determined by the Item identification model and the resolved item class. In some exampies, a response to the alert may be received. The response may include an indication of either the item class determined by the item identification model or the item class resolved from the R FID tag information (e.g., the response may indicate what the item actually was), if the response indicates that the item class determined by the item identification model was correct, the Item identification server 1108 may send an additional aiert (e.g., to alert that other, similar, items may not include RFID tags or include RFID tags that also have errors), if the response indicates that the resolved item class was correct, the item identification server 1108 may provide the image of the item and the (correct) item class as training data to the item identification mode! to retrain the item identification model to better identify similar items.[011S] An item processing system can be implemented in a variety of contexts, such as for managing items in a warehouse, for logistics, at retailer storefronts, at recycling stations, or otherwise.

[0016] In one exampie, an item processing system may be used in a logistics context.RFID tags in the logistics context may be configured to provide logistics-relevant information, such as the material composition, weight, or dimensions, as RFID tag information. A data capture system 1110 for logistics may include additional sensors to capture material composition (e.g,, an infrared scanner), weight (e.g., a scale), and geometric data such as volume and shape (e.g., multiple cameras, LIDAR) and configure a machine learning model to determine an item class based on an image, a material composition, a weight, and geometric data. The item class can include items that require different processes to be shipped., such as lithium batteries or oversized items. The machine learning mode! can be trained to classify items into such different item classes, to facilitate the shipping of various items.

[0117] in another example, a retailer storefront may implement an item processing system at a point-of-sale (POS) terminal, in some situations, the PQS terminal may implement some or all components (excluding the tag 1104) described in FIG. 11. In other situations, the components described in FIG, 11 may be external to the POS terminal, and the POS terminal may be in communication with the components. As an example, the POS terminal may implement the RFID reader system 1102 along with the database 1106 and be communicatively coupled to the item identification server 1108, which may be located in or be remote to the retailer storefront. The retailer may configure RFID tags at the retailer storefront with RFID tag information, such as an EPC, which can be resolved to an item class (e.g., an EPC may directly indicate an item class). The POS terminal may additionally include a camera, which acts as the data capture system 1110. When a customer wishes to purchase an item, the POS terminal may capture an image of the item and obtain the RFID-derived information from the item and, for example, if a determination from an image classifier matches the resolved item class, proceed with the sale. Otherwise, it may alert the retailer of the mismatch. Theretailer may than check if other, similar items may be damaged, or may have damaged or missing RFID tags. Additionally, the item identification may enable the retailer to route items to different processing streams, such as by identifying restricted goods such as alcohol which require additional verification of the age of a customer,fOllS) In yet another example, a recycling facility may implement an item processing system using the components described in FIG, II, Because the recycling process for an item often depends largely on its material composition, an RFID tag associated with the item may be configured to provide the material composition of the Item as RFID tag information. In one such example, the recycling facility may train a vision-based model implemented by the item identification server 1108, such as the image classifier described above, to identify the material composition of Items based on images of the items captured by a camera of the data capture system 1110. The recycling facility may then use the vision-based model to identify the material composition of items to be recycled, and direct items with different compositions to different recycling streams,

[0119] In the examples provided by the figures above, data is depicted and described as flowing directly from one entity to another. This data flow may be facilitated by one or more system interfaces. For example, in FIG. 6, the various interfaces blocks of the reader system 600 are configured to communicate with each other, other interfaces, and other external entities such as the services 632 and 640. In other examples., some of the data may pass from one entity to another through an Interface. Some examples of such communications through interfaces may be the interfaces 608, 618, and 628 configured to communicate with at least the services 632. and 640. Other interfaces, implemented and controlled by various hardware or software components (shown or not shown in the figures) can be configured to receive and send data to and from different entities. For example, an interface can be configured to receive and provide data to various components of an item processing system such as data capture systems, item identification servers, or RFID reader systems. In one example, an interface may receive data about an item from a data capture system and receive corresponding RFID- derived information from an RFID reader system, or a resolver system. The interfacemay then provide the data and the RFiD-derived information to an item identification server and thereafter receive an item class associated with the data and the RFID- derived information from the item identification server.

[0120] In some exampies, an item information system interface can send data such as item classes to external entities, such as to an item or tag manufacturer, an item owner, a fao ly owner or supervisor, a database, an authority, or any suitable entity. The interface or an associated processing block may determine a destination or entity to which the data is to be sent to using any suitable means. For example, the interface may determine the destination from RFID-derived information, from the data, from the item ciass, or from some other information, in some examples, the interface may communicate with an identification system to determine the destination. Such identification systems may include a manufacturer identification system, which is able to identify an item or tag manufacturer, or an owner identification system, which is able to identify an item owner. The interface may provide the RFID-derived information and / or the data such that the identification systems can properly identify the destination for the item class. The manufacturer identification system may additionally provide, or allow access to, supplemental information related to the item and / or the tag, such as characteristics of the item, a manufacturing date of the item, or a TID of the tag. The owner identification system may additionally provide, or allow access to, supplemental information related to the owner of the item or of the history of an item, such as access permissions, item owner history, item owner contact information, or item location history, in one example, the RFID-derived information may include an indication of an entity to which the item class should be sent to, or it may be used to retrieve an indication of an entity, such as by accessing a database using the RFID- derived information. In another example, the data may include an image that can be Used to visually identify an entity, or certain data may be associated with a specification. Similarly, the item class may be associated with a specific entity. In any case, the interface may identify an entity or destination to which the item class is to be sent to.

[0121] TIG. 12 is a flow diagram illustrating an example method for RFID-assisted training of an item identification model using a continual learning process, where the item identification model is trained with RFID-derived information from RFID tags on some items, according to examples.

[0122] Example methods for training of item identification models with supplemental information from RFID tags may include one or more operations, functions or actions as illustrated by one or more of blocks 1222, 1224, and / or 1226. The operations described in blocks 1222 through 1226 may also be stored as computer-executable instructions In a computer-readable medium such as a computer-readable medium 1220 of a computing device 1210.

[0123] A process for training of item identification models with supplemental information from RFID tags may Include block 1222, ''RECEIVE DATA ABOUT AN ITEM AND RECEIVE RFID TAG INFORMATION FROM A TAG ASSOCIATED WITH THE ITEM / ' For example, as depicted in FIG. 7 and FIG. 11, an item processing system may include a data capture system (e.g., data capture system 714, data capture system 1110) and an RFID reader (e.g., RFID reader system 716, RRD reader system 1102). The data capture system may use one or more sensors to capture data about an item and the RFID reader may inventory an RFID tag associated with the item to retrieve RFID tag information. The RFID tag information may include relevant item identification information, or may be resolved by a resolver system or service to obtain the relevant information as described above. The captured data and the RFID-derived information may both be sent to a device implementing the item identification model, such as the server 702 / 802 of FIG, 7 / 8.

[0124] The process may also include block 1224, " TRAIN AN ITEM IDENTIFICATION MODEL? At block 1222, the item identification model may be trained, for example, using a process illustrated by FIG. 8 and FIG. 9. The obtained captured data and RFID- derived information (the RFID tag information and / or information resolved from the RFID tag information) form a data- class pair which can be used to train the item identification model. For example, the data-class pairs can be used by the trainingmodule 824 in training of the learning module 822, implementing the item identification model, in a continual learning process to generate a trained item identification mode!.

[0325] Stock 1224 may be fallowed by block 1226, " FINE-TUNE THE ITEM IDENTIFICATION MODEL." At block 1226, the trained item identification model may be fine-tuned using additional data-class pairs obtained during use of the trained item identification model The additional data-class pairs may be used, for example as depicted in FIG, 9, as validation data to fine-tune the model. The process to obtain validation data is described above with reference to FIG, 11, in which data about an item and corresponding RFID-derived information from a tag attached to the item are received by the item identification server 1108 and are later used for fine-tuning. This fine-tuning of the item Identif ication model may result in a higher performance of item identification model according to one or more metrics, such as the accuracy, precision, or recall as described above,

[0126] The blocks included in the above-described processes are for illustration purposes. RFID-assisted training of item identification models may be performed by similar processes with fewer or additional blocks, in some examples, the blocks may be performed in a different order. In some other examples, various blocks may be eliminated. In still other examples, various blocks may be divided into additional blocks, or combined together into fewer blocks. Although illustrated as sequentially ordered operations, In some implementations the various operations may be performed in a different order, or in some cases various operations may be performed at substantially the same time.

[0127] FIG. 13 illustrates a block diagram of an example computer program product, according to examples. In some examples, as shown in FIG, 13, a computer program product 1300 may include a signal bearing medium 1302 that may also include machine- readable instructions 1304 that, when executed by, for example, a processor, may provide the functionalities described above with respect to FIG. 7 through FIG. 11. Thus, for example, referring to FIG. 8, the item Identification server 802 and the RFID server 804 may undertake one or more of the tasks shown in FIG. 13 in response to themachine-readabie instructions 1304 conveyed to the servers by the signal bearing medium 1302 to perform actions associated with the processes as described herein. Some of those instructions may include receiving RFID tag information from an RFID reader system, training an item identification model using a supervised learning system, and fine-tuning the trained item identification model,

[6128] In some implementations, the signal bearing medium 1302 depicted in H& 13 may encompass a computer-readable medium 1306, such as, but not limited to, a hard disk drive, a Compact Disc (CD), a Digital Versatile Disk (DVD), a digital tape, memory, etc. In some implementations, the signal bearing medium 1302 may encompass a recordable medium 1308, such as, but not limited to, memory, read / write (R / W) CDs, R / W DVDs, etc. In some implementations, the signal bearing medium; 1302 may encompass a communications medium 1310, such as, but not limited to, a digital and / or an analog communication medium (e.g., a fiber optic cable, a waveguide, a wired communications link, a wireless communication link, etc.). Thus, for example, the program product 1300 may be conveyed to one or more modu les of the server 802 by an RF signal bearing medium, where the signal bearing medium 1302 is conveyed by a wireless communications medium 1310 (e,g., a wireless communications medium conforming with the IEEE 802.11 standard).

[0129] According to some examples, a method for an interface to facilitate item identification includes receiving a first item class. In one example, the first item class is obtained from a resolver configured to resolve an item class from an item identifier, the item identifier originating from a radio frequency Identification (RFID) integrated circuit (IC) associated with an item. In another example, the first item class is obtained from a learning system configured to receive an image-class pair, the image-class pair comprising an image of an item and the resolved item class, and to use the image-class pair and a training algorithm to learn an association between the item image and resolved item class. The method also includes determining a destination for the first item class and transmitting the first item class to the destination.

[0130] According to other exampies, the item class indicates one or more item characteristics. The item class may include one or more of a Global Trade item Number (GTIN), an Electronic Product Code (EPC), a Universal Product Code (UPC), or a Stock Keeping Unit (SKU), The item class may not uniquely identify the item. In some examples, determining the destination for the first item classes includes receiving the destination from a manufacturer identification system or an owner identification system. The method may further include sending one or more of item information, RFID IC information, owner Information, or item history information to the destination. The destination may be determined based on one or more of the first item class or the first image. The destination may include one or more, of a database, an external device, or an item owner, In some examples, the algorithm may be a supervised learning algorithm.

[0131] According to some examples, a computing device to execute an item information system includes a communication subsystem configured to communicate with an item identification system, a memory to store instructions, and a processor coupled to the communication subsystem and the memory. The processor can be configured to execute an interface in conjunction with the instructions stored in the memory, and be configured to receive, via the interface, the first item class. The processor may additionally be configured to determine a destination for the first item class, and to transmit the first item class to the destination. The item identification system may include a resolver configured to resolve an item class from an item identifier where the item identifier originates from an RFID IC associated with an item. The item identification system may also include a learning system configured to receive an image¬ class pair, where the image-class pair comprises an image of the Item and the resolved item ciass, and to use the image-class pair and a training algorithm to learn an association between the item image and the resolved item ciass. The item identification system may additionally include a controller configured to receive a first image and if a first item class corresponding to the first image is available, then the controller is configured to send an image-ciass pair formed from the first image and the first item class to the learning system for the learning system to train on using the trainingalgorithm. Otherwise, the controller is configured to send the first image to the learning system for the learning system to provide the first Item class,

[0132] According to other examples, the item class indicates one or more Item characteristics. The item class may include one or more of a Global Trade Item Number (GTINji an Electronic Product Code (EPC), a Universal Product Code (UPC)?or a Stock keeping Unit (SKU), The item class may not uniquely identify the item. In some examples, the processor is configured to determine the destination for the first item class by receiving the destination from a manufacturer identification system or an owner identification system. The processor may be further configured to send one or more of item information, RFID IC information, owner information, or item history information to the destination. The destination may be deter ined based on one or more of the first item class or the first image. The destination may include one or more of a database, an external device, or an item owner. In some examples, the algorithm may be a supervised learning algorithm.

[0133] According to further examples, an item information system comprises an interface, configured to communicate with an item identification system and a controller, and an interface processing block. The interface processing block Is configured to receive, via the interface, the first item class. The interface processing block is also configured to determine a destination for the first item class and to transmit the first item class to the destination. The Item identification system includes a resolver configured to resolve an item class from an item identifier, the item identifier originating from an RFID IC associated with an item. The item identification system also includes a learning system configured to receive an image-class pair, the image-class pair comprising an image of the item and the resolved item class, and also is configured to use the image-class pair and a training algorithm to learn an association between the item image and the resolved item class. The controller is configured to receive a first image and if the first item class corresponding to the first image is available, then to send an image-class pair formed from the first tmageand the first item class to the learning system for the learning system to train on using the training algorithm.Otherwise, the controller is configured to send the first image to the learning system for the learning system to provide the first item class.

[0134] As mentioned previously, examples are directed to RFID-assisted item processing and identification systems. Examples additionally include programs, and methods of operation of the programs. A program is generally defined as a group of steps or operations leading to a desired result, due to the nature of the elements in the steps and their sequence. A program is usually advantageously implemented as a sequence of steps or operations for a processor but may be implemented in other processing elements such as FPGAs, DSPs, or other devices as described above.[0135 Performing the steps, instructions, or operations of a program requires manipulating physical quantities. Usually, though not necessarily, these quantities may be transferred, combined, compared, and otherwise manipulated or processed according to the steps or instructions, and they may also be stored in a computer- readable medium. These quantities include, for example, electrical, magnetic, and electromagnetic charges or particles, states of matter, and in the more general case can include the states of any physical devices or elements. Information represented by the states of these quantities may be referred-to as bits, data bits, samples, values, symbols, characters, terms, numbers, or the like. However, these and similar terms are associated with and merely convenient labels applied to the appropriate physical quantities, individually or in groups.

[0136] Examples furthermore include storage media. Such media, individually or in combination with others, have stored thereon instructions, data, keys, signatures, and other data of a program made according to the examples. A storage medium according to examples is a computer-readable medium, such as a memory, and can be read by a processor of the type mentioned above. If a memory, it can be implemented in any of the ways and using any of the technologies described above.

[0137] Even though it Is said that a program may be stored in a computer-readable medium, it does not need to be a single memory, or even a single machine. Various portions, modules or features of it may reside in separate memories, or even separatemachines. Ths separate machines may be connected directly.; or through a network such as a local access network (LAN) or a global network such as the Internet.

[0138] Often, for the sake of convenience only, it is desirable to implement and describe a program as software. The software can be unitary or thought of in terms of various interconnected distinct software modules,[63.39] The foregoing detailed description has set forth various examples of the devices and / or processes via the use of block diagrams and / or examples. Insofar as such block diagrams and / or examples contain one or more functions and / or aspects, each function and / or aspect within such block diagrams or examples may be implemented individually and / or collectively, by a wide range of hardware, software, firmware, or virtually any combination thereof. Some aspects of the examples disclosed herein, in whole or in part, may be equivalently Implemented employing integrated circuits, as one or more computer programs running on one or more computers (e.g., as one or more programs running on one or more computer systems), as one or more programs running on one or more processors (e.g. as one or more programs running on one or more microprocessors), as firmware, or as virtually any combination thereof, and that designing the circuitry and / or writing the code for the software and / or firmware would be well within the skill of one of skill in the art in light of this disclosure.

[0140] The present disclosure is not to be limited in terms of the particular examples described in this application, which are intended as illustrations of various aspects. Many modifications and variations can be made without departing from its spirit and scope. Functionally equivalent methods and apparatuses within the scope of the disclosure, In addition to those enumerated herein, will be apparent to those skilled in the art from the foregoing descriptions. Such modifications and variations are intended to fall within the scope of the appended claims. The present disclosure is to be limited only by the terms of the appended claims, along with the full scope of equivalents to which such claims are entitled. It is to be understood that this disclosure is not limited to particular methods, configurations, tags, RFICs, readers, systems, and the like, whichcan, of course, vary, ft is also to be understood that the terminology used herein is for the purpose of describing particular examples only, and is not intended to be limiting.

[0141] With respect to the use of substantially any plural and / or singular terms herein, those having skill in the art can translate from the plural to the singular and / or from the singular to the plural as is appropriate to the context and / or application. The various singular / piural permutations may be expressly set forth herein for sake of clarity.

[0142] In general, terms used herein, and especially in the appended claims (e.g., bodies of the appended claims) are generally intended as "open" terms (e.g, the term "including" should be interpreted as "including but not limited to," the term "having" should be interpreted as "having at least," the term "includes" should be interpreted as 'includes but is not limited to," etc.). If a specific number of an introduced claim recitation is intended, such an intent will be explicitly recited in the claim, and in the absence of such recitation no such intent is present. For example, as an aid to understanding, the foliowing appended claims may contain usage of the introductory phrases "at least one" and ’’one or more" to introduce claim recitations. However, the use of such phrases should not be construed to imply that the introduction of a claim recitation by the indefinite articles "a’’ or "an" limits any particular claim containing such introduced claim recitation to examples containing only one such recitation, even when the same claim includes the introductory phrases "one or more” or "at least one" and indefinite articles such as ”a" or "an" (e.g., "a" and / or “an" should be interpreted to mean "at least one" or "one or more"); the same holds true for the use of definite articles used to introduce claim recitations. In addition, even if a specific number of an introduced claim recitation is explicitly recited, such recitation should be interpreted to mean at least the recited number (e.g., the bare recitation of "two recitations," without, other modifiers, means at least two recitations, or two or more recitations).

[0143] Furthermore, In those instances where a convention analogous to "at least one of A, B, and C, etc " is used, in general such a construction is intended in the sense one having skill in the art would understand the convention (e.g., "a system having at least one of A, B, and C" would include but not be limited to systems that have A alone, Balone, C alone, A and 8 together, A and C together, B and C together, and / or A, B, and C together, etc,). Any disjunctive word and / or phrase presenting two or more alternative terms, whether in the description, claims, or drawings, should be understood to contemplate the possibilities of including one of the terms, either of the terms, or both terms. For exampie, the phrase " A or B" will be understood to include the possibilities of " A" or " B" or " A and B."4] For any and all purposes, such as in terms of providing a written description, all ranges disclosed herein also encompass any and ail possible subranges and combinations of subranges thereof. Any listed range can be easily recognized as sufficiently describing and enabling the same range being broken down into at least equal halves, thirds, quarters, fifths, tenths, etc. As a non-limiting example, each range discussed herein can be readily broken down into a lower third, middle third and upper third, etc. All language such as "up to," "at least," "greater than," "less than," and the like include the number recited and refer to ranges which can be subsequently broken down into subranges as discussed above., Finally, a range includes each individual member. Thus, for example, a group having 1-3 cells refers to groups having 1, 2, or 3 cells, Similarly, a group having -5 cells refers to groups having 1, 2, 3, 4, or 5 cells, and so forth.

Claims

Claims1. A computing device associated with an item information system, the computing device comprising:a communication subsystem configured to communicate with an item identification system, wherein the item identification system comprises:a resolver configured to resolve an item class from an item identifier, the item identifier originating from a radio frequency identification (RFID) integrated circuit (IQ associated with an item;a learning system configured to receive an image-class pair, the image-class pair comprising an image of the Item and the resolved item class, and to use the image-class pair and a training algorithm to learn an association between the Item image and the resolved item class; anda controller configured to:receive a first image; andif a first item class corresponding to the first image is available, then send an image-class pair formed from the first image and the first item class to the learning system for the learning system to train on using the training algorithm, else send the first image to the learning system for the learning system to provide the first item class; anda memory to store Instructions; anda processor coupled to the communication subsystem and the memory, the processor to execute an interface in conjunction with the instructions stored In the memory and configured- receive, via the interface, the first item class;determine a destination for the first item class; andtransmit the first item class to the destination.

2. The computing device of claim 1, wherein the item class indicates one or more item characteristics.

3. The computing device of claim 1, wherein the item class includes one or more of a Global Trade item Number GTiN), an Electronic Product Code (EPC), a Universal Product Code (UPC), or a Stock Keeping Unit (SKU).

4. The computing device of claim 1, wherein the item class does not uniquely identify the item..

5. The computing device of claim 1, wherein the processor is configured to determine the destination for the first item class by receiving the destination from a manufacturer identification system or an owner identification system.

6. The computing device of claim 1, wherein the processor is further configured to send one or more of item information, RFID IC information, owner information, or item history in formation to the destination.

7. The computing device of claim 1, wherein the destination is determined based on one or more of the first item class or the first image.

8. The computing device of clai 1, wherein the destination includes one or more of a database, an external device, or an item owner.

9. The computing device of claim 1, wherein the algorithm is a supervised learning algorithm;10. A method for a computing device as described in claims 1-9, the method comprising:receiving the first item class from one or more of the resolver or the learning system;determining the destination; andtransmitting the first item ciass to the determined destination.

Images