System and method for generating hyperspectral artificial vision for machines

Abstract

Embodiments herein provide a method and system for a hyperspectral artificial vision for machines. The system receives a hyperspectral signal of a target material as an input to a neural network model. The system initializes by selecting the number of primitive layers to be used. The system iteratively cycles through all training data (pixels) and updating weights for each unsuccessful material class prediction. Model with two primitives serves as baseline, after which the system adds another primitive layer and repeats the training procedure. The system keeps repeating these processes until obtains convergence. Where the system come to a halt, the system obtains the optimal number of primitives for the given materials. The generated new color pixel is used as a discriminator to aid in locating the target material. The new artificial color is a mixture of weighted chromatic primitives which are optimized for sensitivity / (Spectral Response Functions) SRFs. [To be published with FIG. 2]

Description

TECHNICAL FIELD

[001] The disclosure herein generally relates to the field of machine visionand more specifically, to a method and system for generating a hyperspectralartificial vision for machines.BACKGROUND

[002] Machine vision (MV) is the systems and methods used to provideimaging-based automatic inspection and analysis for applications such as automaticinspection, process control, and robot guidance, usually in industries. Machinevision refers to many technologies, software and hardware products, integratedsystems, actions, methods, and expertise. It attempts to integrate existingtechnologies in new ways and apply them to solve real world problems. Themachine vision is the prevalent one for these functions in industrial automationenvironments but is also used for these functions in other environment vehicleguidance.

[003] A biotic vision uses natural light reflected from an object to extractspatial, temporal, and chemical properties of the object. The spatial properties aregeometric properties of the object such as size and shape, surface texture and so onand the temporal properties are motion of the object. The chemical properties arechemical composition of object or material of the object. The chemical propertiesare commonly acquired using color. Color is a psychological phenomenon whichassociates certain mental sensation (which commonly called as a color) with theenergy reflected by the object. This is by tapping into the energy matter interactionat a molecular level. Depending upon the material and its chemical composition,energy of certain wavelengths is absorbed by the material and the rest is reflected,(some part is scattered away from the direction of the viewer). For example, ripemango would appear red whereas, young would appear green, a dry green leafappears yellow as they reflect both red and green, whereas a fresh leaf appears greenbecause red is absorbed by chlorophyl in the leaf. Thus, color is a discriminatordesigned for separating objects / or detecting objects, especially indicating chemicalcomposition.

[004] Certain animals have more primitives than humans, which haveevolved to help them better perceive their surroundings (i.e., 3 primitives forhumans, 4 primitives for birds, and 16 primitives for mantis shrimps). The adequatenumber of primitives varies depending on the set of materials to be seen by amachine. Usually, color is being used as a discriminator, which can aid in locatingthe target material. To determine the optimal numbers of primitives is practicallyimpossible as this entails changing sensors or filters dynamically during arobot / machine operation.

[005] The existing MV systems typically use filters to capture the light inblue (~400-500 nm) green (~500-600 nm) and red (~600-700 nm) wavelengthrange (called as bands henceforth). The signal from each range is sensed by aCharge Coupled Device (CCD) and the electrical response of the CCD is treated atrichromatic primitive. Though, it shows the scene with colors similar to humanvision, the vision is limited to combination of three primitives. The three primitivesare not necessarily right quantity of primitives and so does the human sensitivityfunctions. Thus, the current mechanism is incapable of enhancing the color orcreating color discriminator for a given task a robot / machine can perform. This isbecause the power distribution spectrum (intensity vs wavelength plot of a reflectedlight from an object) of the object / s is lost in broad band data collection.Resampling such a degraded signal with more primitives is not possible unless anduntil the original power distribution spectrum is retrieved. The hyperspectralimaging techniques uses narrow bands to collect the data, however none of thepresent system use or create a discriminator which is like a color or is a color fordiscriminating objects.SUMMARY

[006] Embodiments of the disclosure present technological improvementsas solutions to one or more of the above-mentioned technical problems recognizedby the inventors in conventional systems. For example, in one embodiment, amethod and system for generating a hyperspectral artificial vision for machines isprovided.

[007] In one aspect, a processor-implemented method for generating ahyperspectral artificial vision for machines is provided. The processor-implemented method comprising receiving via an input / output interface receivinga hyperspectral signal of a target material to be detected as an input to a neuralnetwork model, reshaping each signal vector of the plurality of spectral bands to apredefined shape, dividing the reshaped signal vector of the plurality of spectralbands into a plurality of batches, training the neural network model iteratively foreach of the plurality of batches of the reshaped signal vector to update weight foreach unsuccessful material class prediction, initializing two or more chromaticprimitives of the trained neural network model, optimizing a chromatic primitivesensitivity function using an adaptive moment estimation to achieve optimizedweights of the initialized two or more chromatic primitives, evaluating performanceof the neural network model at each iteration to obtain two or more chromaticprimitives from the plurality of initialized artificial color primitives, generating anew artificial color value for one or more pixels by combining node of the each ofthe obtained two or more chromatic primitives, and predicting an image for thegenerated new color using the learned two or more chromatic primitive sensitivityfunctions to detect the target material.

[008] In another aspect, a system for generating a hyperspectral artificialvision for machines is provided. The system includes an input / output interfaceconfigured to receive a hyperspectral signal of a target material to be detected as aninput to a neural network model, wherein the hyperspectral signal comprising aplurality of spectral bands, one or more hardware processors and at least onememory storing a plurality of instructions, wherein the one or more hardwareprocessors are configured to execute the plurality of instructions stored in the atleast one memory.

[009] Further, the system is configured to reshape each signal vector ofthe plurality of spectral bands to a predefined shape, divide the reshaped signalvector of the plurality of spectral bands into a plurality of batches train the neuralnetwork model iteratively for each of the plurality of batches of the reshaped signalvector to update weight for each unsuccessful material class prediction, initializetwo or more chromatic primitives of the trained neural network model, optimize achromatic primitive sensitivity function using an adaptive moment estimation toachieve optimized weights of the initialized two or more chromatic primitives,evaluate performance of the neural network model at each iteration to obtain twoor more chromatic primitives from the plurality of initialized artificial colorprimitives, generate a new artificial color value for one or more pixels bycombining node of the each of the obtained two or more chromatic primitives, andpredict an image for the generated new color using the learned two or morechromatic primitive sensitivity functions to detect the target material.

[010] In yet another aspect, one or more non-transitory machine-readableinformation storage mediums are provided comprising one or more instructions,which when executed by one or more hardware processors causes a method forgenerating a hyperspectral artificial vision for machines. The processor-implemented method comprising receiving via an input / output interface receivinga hyperspectral signal of a target material to be detected as an input to a neuralnetwork model, reshaping each signal vector of the plurality of spectral bands to apredefined shape, dividing the reshaped signal vector of the plurality of spectralbands into a plurality of batches, training the neural network model iteratively foreach of the plurality of batches of the reshaped signal vector to update weight foreach unsuccessful material class prediction, initializing two or more chromaticprimitives of the trained neural network model, optimizing a chromatic primitivesensitivity function using an adaptive moment estimation to achieve optimizedweights of the initialized two or more chromatic primitives, evaluating performanceof the neural network model at each iteration to obtain two or more chromaticprimitives from the plurality of initialized artificial color primitives, generating anew artificial color value for one or more pixels by combining node of the each ofthe obtained two or more chromatic primitives, and predicting an image for thegenerated new color using the learned two or more chromatic primitive sensitivityfunctions to detect the target material.

[011] It is to be understood that the foregoing general descriptions and thefollowing detailed description are exemplary and explanatory only and are notrestrictive of the invention, as claimed.BRIEF DESCRIPTION OF THE DRAWINGS

[012] The accompanying drawings, which are incorporated in andconstitute a part of this disclosure, illustrate exemplary embodiments and, togetherwith the description, serve to explain the disclosed principles:

[013] FIG. 1 illustrates a block diagram of an exemplary system forgenerating a hyperspectral artificial vision for machines, in accordance with someembodiments of the present disclosure.

[014] FIG. 2 is a flowchart to illustrate the system for generating ahyperspectral artificial vision for machines, in accordance with some embodimentsof the present disclosure.

[015] FIG. 3 is a flowchart to illustrate training of a neural network model,in accordance with some embodiments of the present disclosure.

[016] FIG. 4 is a schematic diagram to illustrate a hyperspectral data cubewith n bands followed by a neural network model architecture, in accordance withsome embodiments of the present disclosure.

[017] FIG. 5 is a schematic diagram depicts a thorough perspective of theneural network architecture, in accordance with some embodiments of the presentdisclosure.

[018] FIG. 6 is a flow diagram to illustrate a method for generating ahyperspectral artificial vision for machines, in accordance with some embodimentsof the present disclosure.

[019] FIG. 7A is a graphical representation of an InternationalCommission on Illumination (CIE) sensitivity function and FIG. 7B is a graphicalrepresentation of CIE filters, in accordance with some embodiments of the presentdisclosure.DETAILED DESCRIPTION OF EMBODIMENTS

[020] Exemplary embodiments are described with reference to theaccompanying drawings. In the figures, the left-most digit(s) of a reference numberidentifies the figure in which the reference number first appears. Whereverconvenient, the same reference numbers are used throughout the drawings to referto the same or like parts. While examples and features of disclosed principles aredescribed herein, modifications, adaptations, and other implementations arepossible without departing from the scope of the disclosed embodiments.

[021] The embodiments herein provide a method and system forgenerating a hyperspectral artificial vision for machines. It should be appreciatedthat a biotic vision uses natural light reflected from the object to extract the spatial,temporal, and chemical properties of the object. Spatial properties of the objectinclude geometric properties such as size and shape, surface texture and so on. Thetemporal properties of the object include objects motion and chemical propertiesinclude chemical composition of object or material of the object. The last propertyis in common sense is a color. Color is a psychological phenomenon whichassociates certain mental sensation (which commonly called as a color) with theenergy reflected by the object. This is by tapping into the energy matter interactionat the molecular level. Depending upon the material and its chemical compositionenergy of certain wavelengths is absorbed by the material and the rest is reflected,(some part is scattered away from the direction of the viewer). For example, ripemango would appear red whereas, young would appear green, a dry green leafappears yellow as they reflect both red and green, whereas a fresh leaf appears greenbecause red is absorbed by chlorophyl in the leaf. Thus, color is a discriminatordesigned for separating objects / or detecting objects, especially indicating chemicalcomposition.

[022] The biotic vision comprising elements such as optical components,sensors, perceptual computation that is color (by layers of natural neural network).Herein, the sensors determine how they respond to the light Humans have threecones responding to short wavelengths (~400 nm to ~500 nm), mediumwavelengths (~500 nm to ~600 nm) and longer wavelengths (~600 nm to ~700 nm)in visible spectrum. They are corresponding to the blue, green, and red sensation.All the colors are seen as result of excitations of these cones in some degree. Theseare not exclusive regions as the response functions peak at certain wavelengths andthe limbs of response function overlap each other. Thus, any color is a sensationresulted combination of excitement of three types of cones. For example, 100%excitation of red and green results in yellow color sensation. Design of sensors andneural processing thus entails deciding number of sensors or primitives, theirspectral response function, positions for peaks (band centers), distance betweenpeaks (band centers) and neural connection to process the signal.

[023] The adequate number of primitives varies depending on the set ofmaterials to be seen by a machine, therefore objective here is to determineappropriate number of primitives as well as spectral response functions (SRFs).Color is being used as a discriminator in this case, which can aid in locating thetarget material. Thus, any sensing system of the camera or robot vision uses onlythree primitives and sensitivity function similar to humans and detect the color. Thecontrol and actuation system then acts on the perceived scene as programed. Forexample, a machine / robot picks up the fruits which are mature and discards all thegreen ones. The machine locates the red object in the scene and directs its gazetoward the same.

[024] Referring now to the drawings, and more particularly to FIG. 1through FIG. 7B, where similar reference characters denote corresponding featuresconsistently throughout the figures, there are shown preferred embodiments andthese embodiments are described in the context of the following exemplary systemand / or method.

[025] FIG. 1 illustrates a block diagram of a system (100) for generating ahyperspectral artificial vision for machines, in accordance with an exampleembodiment. Although the present disclosure is explained considering that thesystem (100) is implemented on a server, it may be understood that the system (100)may comprise one or more computing devices (102), such as a laptop computer, adesktop computer, a notebook, a workstation, a cloud-based computingenvironment and the like. It will be understood that the system (100) may beaccessed through one or more input / output interfaces 104-1, 104-2... 104-N,collectively referred to as I / O interface (104). Examples of the I / O interface (104)may include, but are not limited to, a user interface, a portable computer, a personaldigital assistant, a handheld device, a smartphone, a tablet computer, a workstation,and the like. The I / O interface (104) are communicatively coupled to the system(100) through a network (106).

[026] In an embodiment, the network (106) may be a wireless or a wirednetwork, or a combination thereof. In an example, the network (106) can beimplemented as a computer network, as one of the different types of networks, suchas virtual private network (VPN), intranet, local area network (LAN), wide areanetwork (WAN), the internet, and such. The network (106) may either be adedicated network or a shared network, which represents an association of thedifferent types of networks that use a variety of protocols, for example, HypertextTransfer Protocol (HTTP), Transmission Control Protocol / Internet Protocol(TCP / IP), and Wireless Application Protocol (WAP), to communicate with eachother. Further, the network (106) may include a variety of network devices,including routers, bridges, servers, computing devices, storage devices. Thenetwork devices within the network (106) may interact with the system (100)through communication links.

[027] The system (100) supports various connectivity options such asBLUETOOTH, USB, ZigBee, and other cellular services. The networkenvironment enables connection of various components of the system (100) usingany communication link including Internet, WAN, MAN, and so on. In anexemplary embodiment, the system (100) is implemented to operate as a stand-alone device. In another embodiment, the system (100) may be implemented towork as a loosely coupled device to a smart computing environment. Further, thesystem (100) comprises at least one memory with a plurality of instructions, one ormore databases (112), and one or more hardware processors (108) which arecommunicatively coupled with the at least one memory to execute a plurality ofmodules (114) therein. The components and functionalities of the system (100) aredescribed further in detail.

[028] Herein, the one or more I / O interfaces (104) are configured toreceive a hyperspectral signal of a target material as an input to a neural networkmodel. The hyperspectral signal comprising a plurality of spectral bands. It is to benoted that the hyperspectral signal may come from a small ground sample in caseof a spectrometer data, and it may come as a sequence of signals from a grid ofrows and columns. Further, the system (100) reshapes each signal vector of theplurality of spectral bands to a predefined shape.

[029] Referring FIG. 2, a flow diagram (200), illustrating the system (100)for generating a hyperspectral artificial vision for machines, in accordance withsome embodiments of the present disclosure. The system (100) initializes two ormore chromatic primitives of the neural network model. The two or more chromaticprimitives varies depending on the target material to be seen by a machine. Thesystem (100) initializes by selecting the number of primitive layers to be used. Thismay be done dynamically in the outer flow where the system (100) keeps addingmore primitives as explained above. As an activation function, the system (100)chooses either relu or sigmoid. These hyperparameters are used to define the neuralnetwork model. Further, the system (100) loops for n epochs. Data is divided intobatches for each epoch. The system (100) selects an Adam optimizer, wherein theAdam is an optimization technique that can be used instead of the classicalstochastic gradient descent procedure to update network weights iterative based intraining data. Weights are updated after each batch. Once all of the batches havebeen used for training, the system (100) is configured to compute the loss andmetrics specified during the model compilation. These metrics and weights aresaved, and the model is trained for another epoch.

[030] Further, the system (100) is configured to divide the reshaped signalvector of the plurality of spectral bands into a plurality of batches and trains theneural network model for each of the plurality of batches.

[031] Referring FIG. 3, a flow diagram (300), illustrating the training ofthe neural network model, in accordance with some embodiments of the presentdisclosure. Herein, a hyperspectral data cube to obtain spectral signatures and itsground truth file is being used to get signature labels. The input to the neuralnetwork is a single pixel which has information across all wavelengths. The inputis reshaped to 1 * 1 * n where n denotes the number of bands in the hyperspectralimage. Aim is to discover the best number of primitives (p), and thus starts with p= 2. The system (100) iteratively cycles through all training data (pixels) andupdating weights for each unsuccessful material class prediction. The model withtwo primitives serves as baseline, after which the system (100) adds anotherprimitive layer and repeats the training procedure.

[032] After training, the system (100) is configured to compare it to see ifthere is any significant improvement in the new model over the previous model. Ifthere is growth, the system (100) may continue to add more primitive layers. Thesystem (100) keeps repeating these processes until obtains convergence, i.e., thenew model does not improve. This is where the system (100) come to a halt, andthe system (1000 obtains the optimal number of primitives for the given materials(p-1).

[033] In another example, for a Pavia Centre dataset, the system (100)begins with two primitives. The system (100) continued to add more primitivelayers. However, the neural network model with p = 5 primitives did not show anysignificant improvement over the model with p = 4, so the system stops here andget the optimal number as (p-1 = 4) and SRF's for those four primitives.

[034] Referring FIG. 4, a schematic diagram (400) to illustrate ahyperspectral data cube with n bands followed by the neural network modelarchitecture, in accordance with some embodiments of the present disclosure. Fortraining, the system (100) uses a single pixel at a time, which is represented by greysquares. This input signature is sent to p primitive layers (here p = 3). These are1x1 Convolution layers with a depth of d, where d and n are both equal. Each ofthese layers produces one output. To get new values, the system (100) stacks them.These results are then fed into a fully connected neural network, which predictstheir class.

[035] Further, the system (100) is configured to optimize a chromaticprimitive sensitivity function using an adaptive moment estimation to achieveoptimized weights of the initialized two or more chromatic primitives.

[036] In another embodiment, the system (100) is configured to evaluateperformance of the neural network model at each iteration to obtain two or morechromatic primitives from the plurality of initialized artificial color primitives,wherein halting the training when convergence is obtained.

[037] In the FIG. 5, a schematic diagram (500) depicts a thoroughperspective of a neural network architecture, in accordance with some embodimentsof the present disclosure. As an input, the spectral signatures of n bands are used.These layers are linked to parallel primitive layers of depth d. The architecturepresented above is made up of three primitives. The architecture begins with twoprimitive layers, and the number of layers increases as it converges to an optimalnumber. Each input node is linked to a single primitive layer node. These are logicalconnections, and weights are not learned here. These primitive layers, which aremade up of 1x1 CNN layers, produce a single node output. So, the system (100)gets a connection to a single node from n nodes. This is where the model learnsweights. Weights learned by these primitive layers (1x1 convolution layers) areinterpreted as spectral response functions SRFs. The system (100) creates an SRFfor each primitive layer. These single pixel outputs are now stacked to generate newcolor pixel values. Using pretrained SRF values, these color values can be used togenerate images.

[038] In yet another embodiment, the system (100) is configured togenerate a new artificial color value for one or more pixels by combining node ofthe obtained each of the two or more chromatic primitives, wherein new artificialcolor is a combination of signals created by a linear combination of two or morechromatic primitives. The generated new color pixel is used as a discriminator toaid in locating the target material. The new artificial color is a mixture of weightedchromatic primitives which are optimized for sensitivity / (Spectral ResponseFunctions) SRFs. These Spectral Response Functions (SRFs) can then betransferred to an online deployable system to detect target material. To bettercomplement fresh target content, these SRFs can be updated online, or a newReinforcement Learning (RL) approach may be applied.

[039] The SRFs can be deployed to an embedded machine which has toperform target detection, this type of deployment is done when the machines scanfor different types of materials which are already known to the model as the modelhas already learned SRFs for these materials. The SRFs can also be deployed inreal time systems where the target to be detected is unknown, in this scenario themodel learns new SRFs dynamically by finding the optimal n for a particularcombination of target and data. Learned SRFs performed considerably better thanhuman vision for a given scene / set of materials, and that more primitives may benecessary to better identify target material. The number of primitives varies basedon the materials, as do the SRFs for those primitives.

[040] Further, the system (100) is configured to predict an image for thegenerated new color using the learned two or more chromatic primitive sensitivityfunction to detect the target material. It is to be noted that the trained neural networkmodel enabled with color sensitive functions and two or more chromatic primitivesto run on a neuro-morphic chip, the neuro-morphic chip is enabled with the optimalartificial hyperspectral color vision as an integral part of the machine vision system(100).

[041] Referring FIG. 6, to illustrate a processor-implemented method(600) for generating a hyperspectral artificial vision for machines. Initially, at step(602), receiving, via an input / output interface, a hyperspectral signal of a targetmaterial as an input to a neural network model. The hyperspectral signalcomprising a plurality of spectral bands. The hyperspectral signal may come froma small ground sample in case of a spectrometer data, wherein the hyperspectralsignal come as a sequence of signals from a grid of rows and columns.

[042] At the next step (604), reshaping each signal vector of the pluralityof spectral bands to a predefined shape. The reshaping is essential for preparing thedata so that 2-dimension convolution can be performed on this.

[043] At the next step (606), dividing the reshaped signal vector of theplurality of spectral bands into a plurality of batches. Here, the plurality of spectralbands into different batches as entire data cannot be used to train because of limitedmemory. So, the system determines batch size and data is divided into n batches ofthat size per epoch.

[044] At the next step (608), training the neural network model for eachof the plurality of epochs. At the next step (610), initializing two or more chromaticprimitives of the trained neural network model. At the next step (612), optimizinga chromatic primitive sensitivity function using an adaptive moment estimation toachieve optimized weights of the initialized two or more chromatic primitives. Atthe next step (614), evaluating performance of the neural network model at eachiteration to obtain two or more chromatic primitives from the plurality of initializedartificial color primitives, wherein halting the training when convergence isobtained.

[045] At the next step (616), generating a new artificial color value forone or more pixels by combining node of the obtained each of the two or morechromatic primitives, wherein new artificial color is a combination of signalscreated by a linear combination of two or more chromatic primitives. The two ormore chromatic primitives varies depending on the target material to be seen by amachine. The two or more chromatic primitives are learned optimal chromaticprimitive sensitivity function. The new artificial color is a mixture of weightedchromatic primitives which are optimized for sensitivity / SRF. The generated newcolor pixel is used as a discriminator to aid in locating the target material.

[046] At the next step (618), predicting an image for the generated newcolor using the learned two or more chromatic primitive sensitivity function todetect the target material. Further, the trained neural network model enabled withcolor sensitive functions and two or more chromatic primitives to run on a neuro-morphic chip. The neuro-morphic chip is enabled with the optimal artificialhyperspectral color vision as an integral part of the machine vision system.Experiment:

[047] In first experiment, a neural network architecture capable oflearning color filters similar to human trichromatic vision. The architecture wasdesigned with three parallel 1x1 convolution layers to learn the filters. For thisexperiment, the system (100) used data from the visible region (400nm to 700nm)of the spectral signature. This data is used as an input, and the material label isobtained from the ground truth file. Filters learned from the model as shown inFIG. 7A, a graphical representation, which represents sensitivity function (i.e., anInternational Commission on Illumination (CIE) sensitivity function) hassimilarities to human vision but also distinct. In another experiment, using CIEfilters as shown in FIG. 7B, a graphical representation, for learned filters andcomparing the results to the first experiment. It is discovered that learned filtersoutperformed the model using CIE filters, implying that CIE weights may not beideal. The system (100) is also discovered that each learned primitive is one to onerelated with one of the CIE filters when the system (100) used Spectral AngleMapper (SAM). The table below contains detailed values. Each value in the tableshows the angle (in degrees) between two vectors, and the lower the angle, themore closely the two curves are represented.

[048] The written description describes the subject matter herein to enableany person skilled in the art to make and use the embodiments. The scope of thesubject matter embodiments is defined by the claims and may include othermodifications that occur to those skilled in the art. Such other modifications areintended to be within the scope of the claims if they have similar elements that donot differ from the literal language of the claims or if they include equivalentelements with insubstantial differences from the literal language of the claims.

[049] The embodiments of present disclosure herein address the problemof machine vision. The adequate number of primitives varies depending on the setof materials to be seen by a machine. Usually, color is being used as adiscriminator, which can aid in locating the target material. To determine theoptimal numbers of primitives is practically impossible as this entail changingsensors or filters dynamically during a robot / machine operation. Embodimentsherein provide a method and system for a hyperspectral artificial vision formachines. The system receives a hyperspectral signal of a target material as aninput to a neural network model. The system initializes by selecting the number ofprimitive layers to be used. The system iteratively cycles through all training data(pixels) and updating weights for each unsuccessful material class prediction.Model with two primitives serves as baseline, after which the system adds anotherprimitive layer and repeats the training procedure. The system keeps repeatingthese processes until obtains convergence. Where the system come to a halt, thesystem obtains the optimal number of primitives for the given materials. Thegenerated new color pixel is used as a discriminator to aid in locating the targetmaterial. The new artificial color is a mixture of weighted chromatic primitiveswhich are optimized for sensitivity / (Spectral Response Functions) SRFs.

[050] It is to be understood that the scope of the protection is extended tosuch a program and in addition to a computer-readable means having a messagetherein; such computer-readable storage means contain program-code means forimplementation of one or more steps of the method, when the program runs on aserver or mobile device or any suitable programmable device. The hardware devicecan be any kind of device which can be programmed including e.g., any kind ofcomputer like a server or a personal computer, or the like, or any combinationthereof. The device may also include means which could be e.g., hardware meanslike e.g., an application-specific integrated circuit (ASIC), a field-programmablegate array (FPGA), or a combination of hardware and software means, e.g., anASIC and an FPGA, or at least one microprocessor and at least one memory withsoftware modules located therein. Thus, the means can include both hardwaremeans, and software means. The method embodiments described herein could beimplemented in hardware and software. The device may also include softwaremeans. Alternatively, the embodiments may be implemented on different hardwaredevices, e.g., using a plurality of CPUs.

[051] The embodiments herein can comprise hardware and softwareelements. The embodiments that are implemented in software include but are notlimited to, firmware, resident software, microcode, etc. The functions performedby various modules described herein may be implemented in other modules orcombinations of other modules. For the purposes of this description, a computer-usable or computer readable medium can be any apparatus that can comprise, store,communicate, propagate, or transport the program for use by or in connection withthe instruction execution system, apparatus, or device.

[052] The illustrated steps are set out to explain the exemplaryembodiments shown, and it should be anticipated that ongoing technologicaldevelopment will change the manner in which particular functions are performed.These examples are presented herein for purposes of illustration, and not limitation.Further, the boundaries of the functional building blocks have been arbitrarilydefined herein for the convenience of the description. Alternative boundaries canbe defined so long as the specified functions and relationships thereof areappropriately performed. Alternatives (including equivalents, extensions,variations, deviations, etc., of those described herein) will be apparent to personsskilled in the relevant art(s) based on the teachings contained herein. Suchalternatives fall within the scope of the disclosed embodiments. Also, the words"comprising," "having," "containing," and "including," and other similar forms areintended to be equivalent in meaning and be open ended in that an item or itemsfollowing any one of these words is not meant to be an exhaustive listing of suchitem or items or meant to be limited to only the listed item or items. It must also benoted that as used herein and in the appended claims, the singular forms "a," "an,"and "the" include plural references unless the context clearly dictates otherwise.

[053] Furthermore, one or more computer-readable storage media maybe utilized in implementing embodiments consistent with the present disclosure.A computer-readable storage medium refers to any type of physical memory onwhich information or data readable by a processor may be stored. Thus, acomputer-readable storage medium may store instructions for execution by one ormore processors, including instructions for causing the processor(s) to performsteps or stages consistent with the embodiments described herein. The term"computer-readable medium" should be understood to include tangible items andexclude carrier waves and transient signals, i.e., be non-transitory. Examplesinclude random access memory (RAM), read-only memory (ROM), volatilememory, nonvolatile memory, hard drives, CD ROMs, DVDs, flash drives, disks,and any other known physical storage media.

[054] It is intended that the disclosure and examples be considered asexemplary only, with a true scope of disclosed embodiments being indicated bythe following claims.

Claims

1. A processor-implemented method (600) comprising steps of: receiving (602), via an input / output interface, a hyperspectral signal of a target material to be detected as an input to a neural network model, wherein the hyperspectral signal comprising a plurality of spectral bands; reshaping (604), via one or more hardware processors, each signal vector of the plurality of spectral bands to a predefined shape; dividing (606), via the one or more hardware processors, the reshaped signal vector of the plurality of spectral bands into a plurality of batches; training (608), via the one or more hardware processors, the neural network model iteratively for each of the plurality of batches of the reshaped signal vector to update weight for each unsuccessful material class prediction; initializing (610), via the one or more hardware processors, two or more chromatic primitives of the trained neural network model; optimizing (612), via the one or more hardware processors, a chromatic primitive sensitivity function using an adaptive moment estimation to achieve optimized weights of the initialized two or more chromatic primitives; evaluating (614), via the one or more hardware processors, performance of the neural network model at each iteration to obtain two or more chromatic primitives from the plurality of initialized artificial color primitives, wherein halting the training of the neural network model when convergence of the two or more chromatic primitives is obtained; generating (616), via the one or more hardware processors, a new artificial color value for one or more pixels by combining node of the each of the obtained two or more chromatic primitives, wherein the new artificial color is a combination of signals created by a linear combination of two or more chromatic primitives; and predicting (618), via the one or more hardware processors, an image for the generated new color using the learned two or more chromatic primitive sensitivity functions to detect the target material.

2. The processor-implemented method (600) of claim 1, wherein the trained neural network model is enabled with color sensitive functions and two or more chromatic primitives to run on a neuro-morphic chip.

3. The processor-implemented method (600) of claim 2, wherein the neuro- morphic chip is enabled with an optimal artificial hyperspectral color vision as an integral part of a machine vision system.

4. The processor-implemented method (600) of claim 1, wherein the two or more chromatic primitives varies depending on the target material to be seen by a machine.

5. The processor-implemented method (600) of claim 1, wherein the two or more chromatic primitives are learned optimal chromatic primitive sensitivity function.

6. The processor-implemented method (600) of claim 1, wherein the generated new color pixel is used as a discriminator to aid in locating the target material.

7. The processor-implemented method (600) of claim 1, wherein the hyperspectral signal comes from a ground sample in case of a spectrometer data, wherein the hyperspectral signal come as a sequence of signals from a grid of rows and columns.

8. The processor-implemented method (600) of claim 1, wherein the new artificial color is a mixture of weighted chromatic primitives which are optimized for sensitivity or SRF.

9. A system (100) comprising: an input / output interface (104) to receive a hyperspectral signal of a target material to be detected as an input to a neural network model, wherein the hyperspectral signal comprising a plurality of spectral bands; a memory (110) in communication with the one or more hardware processors (108), wherein the one or more hardware processors (108) are configured to execute programmed instructions stored in the memory (110) to: reshape each signal vector of the plurality of spectral bands to a predefined shape; divide the reshaped signal vector of the plurality of spectral bands into a plurality of batches; train the neural network model iteratively for each of the plurality of batches of the reshaped signal vector to update weight for each unsuccessful material class prediction; initialize two or more chromatic primitives of the trained neural network model; optimize a chromatic primitive sensitivity function using an adaptive moment estimation to achieve optimized weights of the initialized two or more chromatic primitives; evaluate performance of the neural network model at each iteration to obtain two or more chromatic primitives from the plurality of initialized artificial color primitives, wherein halting the training of the neural network model when convergence of the two or more chromatic primitives is obtained; generate a new artificial color value for one or more pixels by combining node of the each of the obtained two or more chromatic primitives, wherein the new artificial color is a combination of signals created by a linear combination of two or more chromatic primitives; and predict an image for the generated new color using the learned two or more chromatic primitive sensitivity functions to detect the target material.

10. A non-transitory computer readable medium storing one or more instructions which when executed by one or more processors on a system, cause the one or more processors to perform method comprising: receiving, via an input / output interface, a hyperspectral signal of a target material as an input to a neural network model, wherein the hyperspectral signal comprising a plurality of spectral bands; reshaping, via one or more hardware processors, each signal vector of the plurality of spectral bands to a predefined shape; dividing, via one or more hardware processors, the reshaped signal vector of the plurality of spectral bands into a plurality of batches; training, via the one or more hardware processors, the neural network model for each of the plurality of batches; initializing, via one or more hardware processors, two or more chromatic primitives of the trained neural network model; optimizing, via the one or more hardware processors, a chromatic primitive sensitivity function using an adaptive moment estimation to achieve optimized weights of the initialized two or more chromatic primitives; evaluating, via the one or more hardware processors, performance of the neural network model at each iteration to obtain two or more chromatic primitives from the plurality of initialized artificial color primitives, wherein halting the training when convergence is obtained; generating, via the one or more hardware processors, a new artificial color value for one or more pixels by combining node of the obtained each of the two or more chromatic primitives, wherein new artificial color is a combination of signals created by a linear combination of two or more chromatic primitives; and predicting, via the one or more hardware processors, an image for the generated new color using the learned two or more chromatic primitive sensitivity function to detect the target material.