System and method for colourizing night-vision forest images

Abstract

The present disclosure introduces a system (102) and method (400) for colorization and classification of night-vision images in forestry environments. The system (102) receives grayscale night-vision images, and through a series of preprocessing steps, including transformation to the Lab color space and regularization, the system constructs a CNN for feature extraction. Training and validation are performed using datasets, and the Lab color space is mapped to the RGB color space to generate the first set of colorized images. Subsequently, YOLOv8 is applied for real-time object detection and classification within the colorized images. The system excels in night vision scenarios, contributing to efficient event detection, wildlife documentation, and forestry management. The integration of YOLOv8 facilitates dynamic decision-making and quick response in forestry environments, making it a valuable tool for comprehensive forestry management and conservation efforts.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the field of image processing. Inparticular, it relates to a system and method for colourization and classification ofnight vision images using Convolutional Neural Networks (CNN) and YOLOv8.BACKGROUND

[0002] Background description includes information that may be useful inunderstanding the present disclosure. It is not an admission that any of theinformation provided herein is prior art or relevant to the presently claimeddisclosure, or that any publication specifically or implicitly referenced is prior art.

[0003] In the realm of black-and-white photo colourization, previous methodshave grappled with challenges such as messy results and a reliance on user input.While attempts have been made to infuse colour into these images, the lack of asystematic classification mechanism has been a significant drawback. Thesemethods, while making strides in adding colour to grayscale images, have fallenshort in providing a systematic and automated approach for classifying the contentwithin these colourized images. The challenge extends beyond the colourizationprocess itself, highlighting a gap in the systematic categorization of the visualelements contained in the images.

[0004] Despite advancements in camera technology, especially in forestenvironments, a significant constraint persists. The prevalent practice involvescapturing night vision photographs or black-and-white images, particularly duringnocturnal operations. This limitation arises even with the utilization of powerfulcameras, emphasizing the need for a more sophisticated and adaptive solution thatcan transcend the constraints of shooting in low-light conditions. The currentmethodologies, therefore, lack the finesse needed to address the nuances of nighttimeimaging within forest settings.

[0005] The inherent limitations of existing techniques in forestry scenariosbecome more pronounced when considering the need for intelligent andsystematic classification of visual elements within the images. The absence ofsuch classification diminishes the utility and insights that can be derived fromthese images, especially in applications such as wildlife monitoring, eventdetection, and ecosystem analysis. As a consequence, there is a clear need for aninnovation that not only enhances the colourization process but also seamlesslyintegrates intelligent classification, overcoming the constraints posed by previousmethodologies and providing a more comprehensive solution for imaging in forestenvironments.

[0006] Therefore, there is a need for a reliable, and robust system and method forcolourization and classification of night vision images accurately.OBJECTS OF THE PRESENT DISCLOSURE

[0007] Some of the objects of the present disclosure, which at least oneembodiment herein satisfies are as listed herein below.

[0008] It is an object of the present disclosure to provide a system and method forcolourization and classification of forestry images, enhancing visual interpretationthrough introduction of colour, improving understanding and analysis.

[0009] It is another object of the present disclosure to provide a system andmethod for colourization and classification of forestry images, incorporatingintelligent object classification using advanced algorithms like CNN andYOLOv8 for accurate identification of elements within ecosystem.

[0010] It is another object of the present disclosure to provide a system andmethod for colourization and classification of forestry images, excelling inscenarios with night vision imagery, converting and classifying black and whiteimages to enable efficient monitoring in low-light conditions.

[0011] It is another object of the present disclosure to provide a system andmethod for colourization and classification of forestry images, contributing toefficient detection of critical events such as fires or unauthorized activities,enhancing response times and mitigation efforts.

[0012] It is another object of the present disclosure to provide a system andmethod for colourization and classification of forestry images, offering a holisticsolution for forestry management, including monitoring animal behaviour,protecting biodiversity, ensuring human safety, and supporting species discovery.SUMMARY

[0013] The present disclosure relates to the field of image processing. Inparticular, it relates to a system and method for colourization and classification ofnight vision images. The system transforms landscape of colourization andclassification for forestry images, enhancing visual interpretation throughintelligent algorithms like CNN and YOLOv8. The system excels in night visionscenarios, contributing to efficient event detection, supporting wildlifedocumentation, and providing a comprehensive solution for forestry management.

[0014] An aspect of the present disclosure pertains to a system that receives a setof grayscale images of night-vision, initiating a pre-processing stage. During thispre-process, the grayscale images are transformed into Lab colour space, andregularization steps are performed. Subsequently, a convolutional neural network(CNN) is constructed, and features are extracted from the pre-processed grayscaleimages, employing a first dataset for training and a second dataset for validation.The transformed Lab colour space is mapped to label RGB colours, generating afirst set of colourized images. The denormalization process follows, ensuringperceptual colour correction to align the colourized images with humanperception. Further, the system applies YOLOv8 for object detection andclassification, resulting in a second set of colourized images.

[0015] Another aspect of the present disclosure pertains to a method forcolourizing night-vision images. The method begins with receiving a set ofgrayscale images, proceeds to pre-processing, transforming them into Lab colourspace, and regularizing the data. A convolutional neural network (CNN) is thenconstructed, extracting features from the pre-processed images through trainingwith a first dataset and validation with a second. The transformed Lab colourspace is mapped to label RGB colours, generating a first set of colourized images,which undergo denormalization. Subsequently, You Look Only Once (YOLOv8)is applied for object detection and classification, resulting in a second set ofcolourized images.

[0016] Various objects, features, aspects, and advantages of the inventive subjectmatter will become more apparent from the following detailed description ofpreferred embodiments, along with the accompanying drawing figures in whichlike numerals represent like components.BRIEF DESCRIPTION OF DRAWINGS

[0017] The accompanying drawings are included to provide a furtherunderstanding of the present disclosure, and are incorporated in, and constitute apart of this specification. The drawings illustrate exemplary embodiments of thepresent disclosure, and together with the description, serve to explain theprinciples of the present disclosure.

[0018] FIG. 1 illustrates an exemplary network architecture of proposed systemfor colourization of night vision images, in accordance with an embodiment of thepresent disclosure.

[0019] FIG. 2 illustrates an exemplary architecture of proposed system forcolourization of night vision images, in accordance with some embodiments of thepresent disclosure.

[0020] FIG. 3A illustrates an exemplary flow chart to illustrate working ofproposed system for colourization of night vision images using CNN, inaccordance with some embodiments of the present disclosure.

[0021] FIG. 3B illustrates an exemplary flow chart to illustrate working ofproposed system for classification of night vision images using YOLOv8, inaccordance with an embodiment of the present disclosure.

[0022] FIG. 4 illustrates an exemplary view of a flow diagram of proposedmethod for colourization of night vision images, in accordance with anembodiment of the present disclosure.

[0023] FIG. 5 illustrates an exemplary computer system in which or with whichembodiments of the present disclosure can be utilized in accordance withembodiments of the present disclosure.DETAILED DESCRIPTION

[0024] The following is a detailed description of embodiments of the disclosuredepicted in the accompanying drawings. The embodiments are in such detail as toclearly communicate the disclosure. However, the amount of detail offered is notintended to limit the anticipated variations of embodiments; on the contrary, theintention is to cover all modifications, equivalents, and alternatives falling withinthe spirit and scope of the present disclosure as defined by the appended claims.

[0025] Embodiment of present disclosure relates to the field of image processing.In particular, it relates to a system and method for colourization and classificationof night vision images.

[0026] An embodiment of the present disclosure pertains to a system that receivesa set of grayscale images of night-vision, initiating a pre-processing stage. Duringthis pre-process, the grayscale images are transformed into Lab colour space, andregularization steps are performed. Subsequently, a convolutional neural network(CNN) is constructed, and features are extracted from the pre-processed grayscaleimages, employing a first dataset for training and a second dataset for validation.The transformed Lab colour space is mapped to label RGB colours, generating afirst set of colourized images. The denormalization process follows, ensuringperceptual colour correction to align the colourized images with humanperception. Further, the system applies YOLOv8 for object detection andclassification, resulting in a second set of colourized images.

[0027] Another embodiment of the present disclosure pertains to a method forcolourizing night-vision images. The method begins with receiving a set ofgrayscale images, proceeds to pre-processing, transforming them into Lab colourspace, and regularizing the data. A convolutional neural network (CNN) is thenconstructed, extracting features from the pre-processed images through trainingwith a first dataset and validation with a second. The transformed Lab colourspace is mapped to label RGB colours, generating a first set of colourized images,which undergo denormalization. Subsequently, You Look Only Once (YOLOv8)is applied for object detection and classification, resulting in a second set ofcolourized images.

[0028] The manner in which the proposed system works, in described in furtherdetails in conjunction with FIGs. 1 to 5. It may be noted that these figure is onlyillustrative, and should not be construed to limit the scope of the subject matter inany manner.

[0029] FIG. 1 illustrates exemplary network architecture 100 of proposed system102 for detecting disease in potato leaves, in accordance with an embodiment ofthe present disclosure.

[0030] In an embodiment, referring to FIG. 1, the system 102 will be connected toa network 104, which is further connected to at least one computing device 108-1,108-2, … 108-N (collectively referred as computing device 108, herein). Thecomputing device 108 may be personal computers, laptops, tablets, wristwatch orany custom-built computing device.

[0031] In an embodiment, the system 102 may receive a set of grayscale images(interchangeably referred to as grayscale images, hereinafter) captured under nightvision conditions within a forest area or a comparable environment by the at leastone computing device 108. These grayscale images are captured under low-lightconditions where conventional visibility may be limited, and specializedequipment or technology, such as night vision devices, cameras or the likes areused to capture these scenes. Those having ordinary skill in the art will understandthat the at least one computing device 108 may be individually referred to ascomputing device 108 and collectively referred to as computing devices 108. Inan embodiment, the computing device 108 may also be referred to as UserEquipment (UE). Accordingly, the terms "computing device" and "UserEquipment" may be used interchangeably throughout the disclosure.

[0032] In an embodiment, the computing device 108 may transmit the imagesover a point-to-point or point-to-multipoint communication channel or network104 to the system 102.

[0033] In an exemplary embodiment, the communication network 104 mayinclude, but not be limited to, a wireless network, a wired network, an internet, anintranet, a public network, a private network, a packet-switched network, a circuitswitchednetwork, an ad hoc network, an infrastructure network, a Public-Switched Telephone Network (PSTN), a cable network, a cellular network, asatellite network, a fiber optic network, or some combination thereof.

[0034] In an embodiment, the computing devices 108 may communicate with thesystem 102 via a set of executable instructions residing on any operating systemto receive the grayscale images. In an embodiment, the one or more computingdevices 108 may include, but not be limited to, any electrical, electronic, electromechanical,or an equipment, or a combination of one or more of the abovedevices such as mobile phone, smartphone, Virtual Reality (VR) devices,Augmented Reality (AR) devices, laptop, a general-purpose computer, desktop,personal digital assistant, tablet computer, mainframe computer, or any othercomputing device. It may be appreciated that the one or more computing devices108 may not be restricted to the mentioned devices and various other devices maybe used.

[0035] In an embodiment, the network 104 is further configured with acentralized server 110 including a database. The centralized server 110 acts as ahub, equipped with a database, facilitating organized storage and retrieval ofcritical information. The centralized server enhances the effectiveness of thenetwork by providing a centralized point for data management and processing,essential for the successful implementation of the described colourization andclassification system.

[0036] In an embodiment, the system 102 pre-process the received grayscaleimages that are transformed into Lab colour space, and regularization steps areperformed. Subsequently, a convolutional neural network (CNN) is constructed,and features are extracted from the pre-processed grayscale images, employing afirst dataset for training and a second dataset for validation. The transformed Labcolour space is mapped to label RGB colours, generating a first set of colourizedimages. The denormalization process follows, ensuring perceptual colourcorrection to align the colourized images with human perception. Further, thesystem 102 applies YOLOv8 for object detection and classification, resulting in asecond set of colourized images.

[0037] Although FIG. 1 shows exemplary components of the network architecture100, in other embodiments, the network architecture 100 may include fewercomponents, different components, differently arranged components, or additionalfunctional components than depicted in FIG. 1. Additionally, or alternatively, oneor more components of the network architecture 100 may perform functionsdescribed as being performed by one or more other components of the networkarchitecture 100.

[0038] FIG. 2 illustrates an exemplary architecture of proposed system forcolourization of night vision images, in accordance with some embodiments of thepresent disclosure.

[0039] In an aspect, referring to FIG. 2, the system 102 may comprise one ormore processor(s) 202. The one or more processor(s) 202 may be implemented asone or more microprocessors, microcomputers, microcontrollers, edge or fogmicrocontrollers, digital signal processors, central processing units, logiccircuitries, and / or any devices that process data based on operational instructions.Among other capabilities, the one or more processor(s) 202 may be configured tofetch and execute computer-readable instructions stored in a memory 204 of thesystem 102. The memory 204 may be configured to store one or more computerreadableinstructions or routines in a non-transitory computer readable storagemedium, which may be fetched and executed to create or share data packets over anetwork service. The memory 204 may comprise any non-transitory storagedevice including, for example, volatile memory such as Random Access Memory(RAM), or non-volatile memory such as Erasable Programmable Read-OnlyMemory (EPROM), flash memory, and the like.

[0040] Referring to FIG. 2, the system 102 may include an interface(s) 206. Theinterface(s) 206 may comprise a variety of interfaces, for example, interfaces fordata input and output devices, referred to as I / O devices, storage devices, and thelike. The interface(s) 206 may facilitate communication to / from the system 102.The interface(s) 206 may also provide a communication pathway for one or morecomponents of the system 102. Examples of such components include, but are notlimited to, processing unit / engine(s) 208 and a local database 210.

[0041] In an embodiment, the processing unit / engine(s) 208 may be implementedas a combination of hardware and programming (for example, programmableinstructions) to implement one or more functionalities of the processing engine(s)208. In examples described herein, such combinations of hardware andprogramming may be implemented in several different ways. For example, theprogramming for the processing engine(s) 208 may be processor-executableinstructions stored on a non-transitory machine-readable storage medium and thehardware for the processing engine(s) 208 may comprise a processing resource(for example, one or more processors), to execute such instructions. In the presentexamples, the machine-readable storage medium may store instructions that, whenexecuted by the processing resource, implement the processing engine(s) 208. Insuch examples, the system 102 may comprise the machine-readable storagemedium storing the instructions and the processing resource to execute theinstructions, or the machine-readable storage medium may be separate butaccessible to the system 102 and the processing resource. In other examples, theprocessing engine(s) 208 may be implemented by electronic circuitry.

[0042] In an embodiment, the local database 210 may comprise data that may beeither stored or generated as a result of functionalities implemented by any of thecomponents of the processor 202 or the processing engines 208. In anembodiment, the local database 210 may be separate from the system 102.

[0043] In an exemplary embodiment, the processing engine 208 may include oneor more engines selected from any of a receiving module 212, a pre-processingmodule 214, a convolutional neural network (CNN) building module 216, afeature extraction module 218, a training and testing module 220, a postprocessingmodule 222, an object detection and classification module 224, anoutput module 226, and other modules 228 having functions that may include butare not limited to testing, storage, and peripheral functions, such as wirelesscommunication unit for remote operation, audio unit for alerts and the like.

[0044] In an embodiment, the receiving module 212 is configured to receive a setof grayscale images of night-vision of an area from one or more devices (notshown). For instance, cameras or sensors strategically positioned throughout theforest may capture grayscale images of the surroundings, particularly in night 30vision scenarios. These devices could include surveillance cameras, specializednight-vision cameras, or other imaging sensors designed for environmentalmonitoring.

[0045] In an embodiment, the pre-processing module 214 is configured to executea series of operations on the received set of grayscale images. Firstly, thegrayscale images are transformed to a Lab colour space image data. The Labcolour space is a colour model that separates image information into threechannels: L* (luminance), a* (green to red), and b* (blue to yellow). Thistransformation may help in extracting more meaningful features from the imagesfor subsequent processing.

[0046] Additionally, a regularization process is applied during pre-processing.This regularization involves several steps. Firstly, the pre-processing moduleadjusts the pixel intensities of the transformed Lab colour space image data. Thisadjustment may include normalizing or enhancing the pixel values to optimize theimage for further analysis. Further, spatial smoothing is applied to the transformedLab colour space image data. Spatial smoothing involves reducing noise orvariations in pixel values across the image, which can contribute to a cleaner andmore consistent representation. Furthermore, contrast normalization isimplemented during regularization. The pre-processing module 214 enhancesvisibility of features within the images by adjusting the contrast levels, making thesubsequent colourization and classification processes more effective.

[0047] In an embodiment, the CNN building module 216 is configured toconstruct a convolutional neural network (CNN). This include defining thearchitecture of the neural network, including the number of layers, types of layers(e.g., convolutional layers, pooling layers), and how these layers areinterconnected. In an exemplary embodiment, by constructing CNN, the systemleverage capabilities of deep learning for extracting intricate features and patternsfrom the pre-processed set of grayscale images. The hierarchical structure ofCNNs allows them to learn progressively more complex representations asinformation passes through the network.

[0048] In an embodiment, the feature extraction module 218 is configured toextract one or more features from the pre-processed set of grayscale images usingthe constructed CNN. The features extracted from the pre-processed set ofgrayscale images include temporal features, capturing dynamic information innight-vision scenes. These temporal features are likely to represent changes andmovements occurring over time, providing the system with the ability to discerndynamic elements within the images, such as animal movements or other temporalvariations. Additionally, the features may include any or a combination of pattern,edge, and texture. By including temporal features and a variety of visualcharacteristics, the feature extraction module enhances the system's ability tocomprehend and interpret the content of the grayscale images, which is vital forsubsequent stages of processing and analysis.

[0049] In an embodiment, the training and testing module 220 is configured totrain the CNN using a first dataset (i.e. training dataset), and validate the firstdataset by utilizing a second dataset (i.e. testing dataset). For an instance, trainingincludes presenting the CNN with a large set of input data (images) along withtheir corresponding expected outputs. The CNN adjusts its internal parametersthrough iterative optimization processes to learn and generalize patterns from thetraining dataset. Further, to ensure the trained CNN's effectiveness andgeneralizability, the module further validates its performance using a seconddataset, known as the testing dataset. The testing dataset is distinct from thetraining dataset and comprises images that the CNN has not encountered duringthe training phase.

[0050] In an embodiment, the post-processing module 222 is configured to mapthe transformed Lab colour space image data by a label RGB colour to generate afirst set of colourized images. The Lab colour space is a colour model thatrepresents colour information separately for intensity (L*) and colour information(a*, b*). Mapping to RGB involves converting these Lab colour values tostandard RGB colour values, which are commonly used for digital images. Aftermapping to RGB colours, the post-processing module 222 denormalizes the firstset of colourized images. The denormalization is a process that reverses thenormalization steps applied during pre-processing. It ensures that the colourizedimages are restored to their original perceptual appearance, accounting for anyadjustments made during normalization. Further, the post-processing module 222includes a perceptual colour correction step. This step is crucial for aligning thecolourized set of images with human perception. It adjusts the colour values toensure that the colourized images appear visually accurate and natural to humanobservers.

[0051] In an embodiment, the object detection and classification module 224 isconfigured to apply You Look Only Once (YOLOv8) on the first set of colourizedimages received from the denormalization process for object detection andclassification, and consequently receive a second set of colourized image. Forobject detection and classification from the first set of colourized images, thepost-processing module 222 is configured to divide the first set of colourizedimages to generate a grid of cells for object localization, this grid aids in preciselylocalizing objects within the images. Additionally, the post-processing module222 defines bounding boxes with pre-set anchor boxes, utilize one or moreconvolutional layers to process and extract features from the first set of colourizedimages. These bounding boxes serve to encapsulate and precisely identify thelocation of objects in the images. Additionally, apply one or more filters to theone or more convolutional layers to collect varied visual patterns, and integrateone or more detection layers to predict bounding boxes and class probabilities.These detection layers play a crucial role in identifying the presence of objectsand assigning them to specific classes.

[0052] Further, the post-processing module 222 modifies the anticipatedbounding boxes in relation to the grid cells position and predetermined anchorboxes, calculates an objectness score to identify and remove portions of thecolourized images that do not contain objects. Furthermore, the post-processingmodule 222 obtains class probabilities for each bounding box, predicts theprobability distribution of each bounding box within predeterminedclassifications; and generate the second set of colourized image includingbounding boxes related class labels, and corresponding confidence ratings,providing a detailed classification of objects in the colourized images.

[0053] In an embodiment, the output module 226 is configured to display thegenerated second set of colourized image to a display device or a computingdevice 108 having an in-build display device.

[0054] FIG. 3A illustrates flow chart to illustrate working of proposed system forcolourization of night vision images using CNN, in accordance with someembodiments of the present disclosure.

[0055] As illustrated in FIG. 3A, a flow chart 300 is disclosed. At step 302,grayscale images are received as input, and pre-processing is performed by aprocessor (202) at step 304. The pre-processing step includes transforming thereceived images to colour spaces and regularizing the images, as shown in steps306 and 308, respectively. Subsequently, at step 310, a CNN is constructed, andfeatures are extracted at step 312. Further, training is conducted at step 314, andtesting is performed at step 316. Furthermore, post-processing is carried out atstep 318, which includes labelling in the RGB colour space and denormalizing theimages at steps 320 and 322, respectively. The colour image is then obtained as anoutput using the CNN at step 324.

[0056] FIG. 3B illustrates an exemplary flow chart 330 to illustrate working ofproposed system for classification of night vision images using YOLOv8, inaccordance with an embodiment of the present disclosure.

[0057] As depicted in FIG. 3B, output generated at step 324 serves as the inputimage for the subsequent image classification step at 332. In this process, pre-setanchor boxes are defined at step 334. Following this, one or more convolutionallayers are employed in step 336 to process and extract features from the receivedinput image. The integration of one or more detection layers occurs in steps 336,338, and 340, predicting both bounding boxes and class probabilities. The finaloutcome of this series of steps is an output image, i.e. colour image in step 342.

[0058] FIG. 4 illustrates an exemplary view of a flow diagram of proposedmethod for colourization of night vision images, in accordance with anembodiment of the present disclosure.

[0059] In an embodiment, a method 400 for colourization of night vision imagesis disclosed. At step 402, a processor 202, receiving a set of grayscale images ofnight-vision of forest or any other area from camera or similar devices attached tothe area. At step 404, the processor 202 pre-processing, the received set ofgrayscale images, and during pre-process the received set of grayscale images aretransformed to a lab colour space image data, and regularization being performed.The regularization step, further including adjusting pixel intensities of thetransformed Lab colour space image data, applying spatial smoothing to thetransformed Lab colour space image data, and implementing contrastnormalization to the transformed Lab colour space image data.

[0060] At step 406, the processor 202 constructing a convolutional neuralnetwork (CNN), and extract one or more features from the pre-processed set ofgrayscale images using the constructed CNN. The features extracted from the preprocessedset of grayscale images comprise temporal features, capturing dynamicinformation in night-vision scenes. Further, the features include any or acombination of pattern, edge, and texture.

[0061] At step 408, the processor 202 training the CNN using a first dataset (i.e.training dataset), and validate the first dataset by utilizing a second dataset (i.e.testing dataset).

[0062] At step 410, the processor 202, mapping the transformed Lab colour spaceimage data by a label RGB colour to generate a first set of colourized images, anddenormalizing the first set of colourized images. The denormalizing step includesa perceptual colour correction step to align the colourized set of images withhuman perception.

[0063] At step 412, the processor 202 applying You Look Only Once (YOLOv8)on the first set of colourized images received from the denormalization process forobject detection and classification, and consequently receive a second set ofcolourized image. The method further includes the following steps: dividing thefirst set of colourized images to generate a grid of cells for object localization,defining bounding boxes with pre-set anchor boxes, utilizing one or moreconvolutional layers to process and extract features from the first set of colourizedimages, and applying one or more filters to the convolutional layers to collectvaried visual patterns. Integration of one or more detection layers follows topredict bounding boxes and class probabilities. The method also includesmodifying the anticipated bounding boxes based on the grid cells' position andpredetermined anchor boxes, calculating an objectness score to identify andremove portions of the first set of colourized images devoid of objects, obtainingclass probabilities for each bounding box, predicting the probability distributionof each bounding box within predetermined classifications, and generating thesecond set of colourized images, including bounding boxes with related classlabels and corresponding confidence ratings.

[0064] FIG. 5 illustrates an exemplary computer system in which or with whichembodiments of the present disclosure can be utilized in accordance withembodiments of the present disclosure.

[0065] Referring to FIG. 5, computer system includes an external storage device510, a bus 520, a main memory 530, a read only memory 540, a mass storagedevice 550, communication port 560, and a processor 570. Those having skilled inthe art will appreciate that computer system may include more than one processorand communication ports. Examples of processor 570 include, but are not limitedto, an Intel Itanium or Itanium 2 processor(s), or AMD Opteron or AthlonMP processor(s), Motorola lines of processors, FortiSOC system on a chipprocessors or other future processors. Processor 570 may include various modulesassociated with embodiments of the present disclosure. Communication port 560can be any of an RS-232 port for use with a modem based dialup connection, a10 / 100 Ethernet port, a Gigabit or 10 Gigabit port using copper or fiber, a serialport, a parallel port, or other existing or future ports. Communication port 560may be chosen depending on a network, such a Local Area Network (LAN), WideArea Network (WAN), or any network to which computer system connects.

[0066] In an embodiment, the memory 530 can be Random Access Memory(RAM), or any other dynamic storage device commonly known in the art. Readonly memory 540 can be any static storage device(s) e.g., but not limited to, aProgrammable Read Only Memory (PROM) chips for storing static informatione.g., start-up or BIOS instructions for processor 570. Mass storage 560 may beany current or future mass storage solution, which can be used to storeinformation and / or instructions. Exemplary mass storage solutions include, but arenot limited to, Parallel Advanced Technology Attachment (PATA) or SerialAdvanced Technology Attachment (SATA) hard disk drives or solid-state drives(internal or external, e.g., having Universal Serial Bus (USB) and / or Firewireinterfaces), e.g. those available from Seagate (e.g., the Seagate Barracuda 7102family) or Hitachi (e.g., the Hitachi Deskstar 7K1000), one or more optical discs,Redundant Array of Independent Disks (RAID) storage, e.g. an array of disks(e.g., SATA arrays), available from various vendors including Dot Hill SystemsCorp., LaCie, Nexsan Technologies, Inc. and Enhance Technology, Inc.

[0067] In an embodiment, the bus 520 communicatively couples processor(s) 570with the other memory, storage and communication blocks. Bus 520 can be, e.g. aPeripheral Component Interconnect (PCI) / PCI Extended (PCI-X) bus, SmallComputer System Interface (SCSI), USB or the like, for connecting expansioncards, drives and other subsystems as well as other buses, such a front side bus(FSB), which connects processor 570 to software system.

[0068] In another embodiment, operator and administrative interfaces, e.g. adisplay, keyboard, and a cursor control device, may also be coupled to bus 520 tosupport direct operator interaction with computer system. Other operator andadministrative interfaces can be provided through network connections connectedthrough communication port 560. External storage device 510 can be any kind ofexternal hard-drives, floppy drives, IOMEGA Zip Drives, Compact Disc - ReadOnly Memory (CD-ROM), Compact Disc - Re-Writable (CD-RW), Digital VideoDisk - Read Only Memory (DVD-ROM). Components described above are meantonly to exemplify various possibilities. In no way should the aforementionedexemplary computer system limit the scope of the present disclosure.

[0069] Thus, the present disclosure provides a system and method torevolutionize colourization and classification of forestry images, introducingenhanced visual interpretation through intelligent algorithms like CNN andYOLOv8. While the foregoing describes various embodiments of the disclosure,other and further embodiments of the disclosure may be devised without departingfrom the basic scope thereof. The scope of the disclosure is determined by theclaims that follow. The disclosure is not limited to the described embodiments,versions or examples, which are included to enable those having ordinary skill inthe art to make and use the disclosure when combined with information andknowledge available to the person having ordinary skill in the art.ADVANTAGES OF THE PRESENT DISCLOSURE

[0070] The present disclosure provides a system and method for colourization andcategorization of images within forestry environments.

[0071] The present disclosure provides a system and method for colourization andclassification of forestry images, enhancing the interpretation of visuals byincorporating colour, thereby improving comprehension and analysis.

[0072] The present disclosure provides a system and method for colourization andclassification of forestry images, integrating advanced algorithms such as CNNand YOLOv8 for intelligent object classification, ensuring precise identificationof elements within the ecosystem.

[0073] The present disclosure provides a system and method for colourization andclassification of forestry images, excelling in scenarios with night vision imagery,this includes involves converting and categorizing black and white images toenable effective monitoring in low-light conditions.

[0074] The present disclosure provides a system and method for colourization andclassification of forestry images, contributing to the efficient detection of criticalevents such as fires or unauthorized activities that contributes to enhancedresponse times and mitigation efforts.

Claims

1. A system (102) for colourization and classification of night-vision images, the system (102) comprising: one or more processors (202) coupled with a memory (204), wherein said memory (204) stores instructions which when executed by the one or more processors (202) causes the system (102) to: receive a set of grayscale images of night-vision; pre-process the received set of grayscale images, wherein during pre-process the received set of grayscale images are transformed to a lab colour space image data, and regularization being performed; construct a convolutional neural network (CNN), and extract one or more features from the pre-processed set of grayscale images using the constructed CNN; train the CNN using a first dataset, and validate the first dataset by utilizing a second dataset; map the transformed Lab colour space image data by a label RGB colour to generate a first set of colourized images, and denormalize the first set of colourized images; and apply You Look Only Once (YOLOv8) on the first set of colourized images received from the denormalization process for object detection and classification, and consequently receive a second set of colourized image.

2. The system (102) as claimed in claim 1, wherein the one or more features extracted from the pre-processed set of grayscale images comprise temporal features, capturing dynamic information in night-vision scenes.

3. The system (102) as claimed in claim 1, wherein the one or more features comprise any or a combination of pattern, edge, and texture.

4. The system (102) as claimed in claim 1, wherein the denormalization comprise a perceptual colour correction step to align the colourized set of images with human perception.

5. The system (102) as claimed in claim 1, wherein during regularization, the system is configured to: adjust pixel intensities of the transformed Lab colour space image data; apply spatial smoothing to the transformed Lab colour space image data; and implement contrast normalization to the transformed Lab colour space image data.

6. The system (102) as claimed in claim 1, wherein for object detection and classification from the first set of colourized images, the system is configured to: divide the first set of colourized images to generate a grid of cells for object localization; define bounding boxes with pre-set anchor boxes; utilize one or more convolutional layers to process and extract features from the first set of colourized images, and apply one or more filters to the one or more convolutional layers to collect varied visual patterns; integrate one or more detection layers to predict bounding boxes and class probabilities; modify the anticipated bounding boxes in relation to the grid cells position and predetermined anchor boxes; calculate objectness score to identify and remove portions of the first set of colourized images devoid of objects; obtain class probabilities for each bounding box; predict the probability distribution of each bounding box within predetermined classifications; and generate the second set of colourized image including bounding boxes related class labels, and corresponding confidence ratings.

7. A method (400) for colourizing and classifying night-vision images, the method comprising the steps of: receiving (402), by a processor (202), a set of grayscale images of night-vision; pre-processing (404), the received set of grayscale images, by the processor (202), wherein during pre-process the received set of grayscale images are transformed to a lab colour space image data, and regularization being performed; constructing (406), by the processor (202), a convolutional neural network (CNN), and extract one or more features from the pre-processed set of grayscale images using the constructed CNN; training (408), by the processor (202), the CNN using a first dataset, and validate the first dataset by utilizing a second dataset; mapping (410), by the processor (202), the transformed Lab colour space image data by a label RGB colour to generate a first set of colourized images, and denormalizing the first set of colourized images; and applying (412), by the processor (202), You Look Only Once (YOLOv8) on the first set of colourized images received from the denormalization process for object detection and classification, and consequently receive a second set of colourized image.

8. The method (400) as claimed in claim 7, wherein the denormalization step comprising a perceptual colour correction step to align the colourized set of images with human perception.

9. The method (400) as claimed in claim 7, wherein regularization step, further comprising: adjusting pixel intensities of the transformed Lab colour space image data; applying spatial smoothing to the transformed Lab colour space image data; and implementing contrast normalization to the transformed Lab colour space image data.

10. The method (400) as claimed in claim 7, wherein for object detection and classification from the first set of colourized images, the method comprises steps of: dividing the first set of colourized images to generate a grid of cells for object localization; defining bounding boxes with pre-set anchor boxes; utilizing one or more convolutional layers to process and extract features from the first set of colourized images, and apply one or more filters to the one or more convolutional layers to collect varied visual patterns; integrating one or more detection layers to predict bounding boxes and class probabilities; modifying the anticipated bounding boxes in relation to the grid cells position and predetermined anchor boxes; calculating objectness score to identify and remove portions of the first set of colourized images devoid of objects; obtaining class probabilities for each bounding box; predicting the probability distribution of each bounding box within predetermined classifications; and generating the second set of colourized image including bounding boxes related class labels, and corresponding confidence ratings.