Device and method of classifying bloodstain pattern using deep learning

Abstract

A method for classifying bloodstain shapes using deep learning according to one embodiment of the present invention comprises: acquiring a plurality of bloodstain images; classifying and labeling the acquired bloodstain images based on a predefined classification system; storing the classified and labeled images to create a bloodstain image database; extracting training data and validation data from the bloodstain image database and separating them according to a pre-set ratio; training a deep learning model using the training data; verifying the trained deep learning model using the validation data; and classifying the shapes of the bloodstain images using the trained deep learning model. The predefined classification system may include a classification system based on the generation mechanism of bloodstains or a morphological classification system of bloodstains.

Description

Technology Field

[0001] Embodiments of the present invention relate to a bloodstain shape classification device and method using deep learning. Background Technology

[0002] A bloodstain is a trace formed when blood falls, splashes, or spreads on a surface or object. Bloodstains are not merely simple stains; they appear in various forms depending on the manner in which the blood flowed and environmental factors at the time.

[0003] In criminal cases, Bloodstain Pattern Analysis (BPA) plays an important role in reconstructing the scene of a bloody incident to identify the crime scene.

[0004] Bloodstain pattern analysis is conducted according to methodologies of distinguishing and observing bloodstain patterns, classifying bloodstain patterns, identifying bloodstain behavior, and estimating and reconstructing behavior. Determining human behavior using these bloodstain pattern analysis techniques requires extensive experience and specialized knowledge.

[0005] Classification of bloodstain types is one of the most important steps in the bloodstain type analysis process, and classifying and distinguishing various bloodstain types is a core task in the field of bloodstain type analysis.

[0006] Scene reconstruction using bloodstain patterns is carried out by organizing the morphological characteristics of bloodstains formed by specific actions, classifying the bloodstains remaining at the scene through this, and inversely estimating the actions required to create each bloodstain.

[0007] Currently, the analysis of bloodstain patterns is carried out through the knowledge and analysis of experts throughout the entire process, and this process has the problem of requiring excessive manpower and time for accurate analysis.

[0008] By utilizing deep learning technology, feature points within an image can be accurately distinguished, which enables faster and more precise analysis of bloodstain patterns. The problem to be solved

[0009] Embodiments of the present invention aim to provide a bloodstain shape classification device and method using deep learning, which reduces expert fatigue, increases analysis speed, and improves accuracy during on-site analysis through the automatic classification of complex bloodstains. means of solving the problem

[0010] A method for classifying bloodstain shapes using deep learning according to one embodiment of the present invention comprises: acquiring a plurality of bloodstain images; classifying and labeling the acquired bloodstain images based on a predefined classification system; storing the classified and labeled images to create a bloodstain image database; extracting training data and validation data from the bloodstain image database and separating them according to a pre-set ratio; training a deep learning model using the training data; verifying the trained deep learning model using the validation data; and classifying the shapes of the bloodstain images using the trained deep learning model. The predefined classification system may include a classification system based on the generation mechanism of bloodstains or a morphological classification system of bloodstains.

[0011] In one embodiment of the present invention, the bloodstain image may include an original bloodstain image and an extended bloodstain image.

[0012] In one embodiment of the present invention, the expanded bloodstain image can be generated by dividing the original bloodstain image into a plurality of regions and applying an opaque filter to the divided bloodstain images.

[0013] In one embodiment of the present invention, the operation of outputting the shape classification result of the bloodstain image may be further included.

[0014] In one embodiment of the present invention, based on the verification result of the trained deep learning model, the method may include the operation of performing additional training of the trained deep learning model using additional bloodstain images.

[0015] In one embodiment of the present invention, the classification system based on the generation mechanism of the bloodstain comprises: a first-1 classification classifying the bloodstain into scattering bloodstains and non-scattering bloodstains; a first-2 classification classifying the scattering bloodstains according to the generation mechanisms of gravitational action, external force dispersion, vascular pressure ejection, and detachment, and the non-scattering bloodstains according to the generation mechanisms of surface action and transference; a second classification classifying the bloodstains caused by gravitational action into falling bloodstains, falling connected bloodstains, and accumulated bloodstains; a bloodstain caused by external force dispersion into impact scattering bloodstains and exhalation bloodstains; a bloodstain caused by vascular pressure ejection into linear ejection bloodstains, massive ejection bloodstains, massive outflow bloodstains, and arterial non-bloodstains; a bloodstain caused by detachment into detachment bloodstains; a bloodstain caused by surface action into pooled bloodstains, flow bloodstains, and absorption bloodstains; and a bloodstain caused by transference The bloodstains may include a 1-3 classification of contact bloodstains, and a 1-4 classification in which the impact scattering bloodstains are classified into blunt force gunshot scattering bloodstains and gunshot impact scattering bloodstains, the linear spurting bloodstains are classified into wrist linear spurting bloodstains and carotid artery linear spurting bloodstains, the detaching bloodstains are classified into swinging detaching bloodstains and stopping detaching bloodstains, the stopping detaching bloodstains are classified into front stopping detaching bloodstains and rear stopping detaching bloodstains, the contact bloodstains are classified into shape-transitioning bloodstains, irregular shape-transitioning bloodstains, and rubbing bloodstains, and the rubbing bloodstains are classified into wiped bloodstains and buried bloodstains.

[0016] In one embodiment of the present invention, the morphological classification system of the bloodstain comprises: a 2-1 classification classifying the bloodstain into scatter bloodstains and non-scattering bloodstains according to bloodstain morphological characteristics; a 2-2 classification classifying the scatter bloodstains into linear scatter bloodstains and non-linear scatter bloodstains according to whether there is a linear distribution, and classifying the non-scattering bloodstains into irregular boundary bloodstains and regular boundary bloodstains according to whether there is regularity of the boundary; classifying the linear scatter bloodstains into linear ejection bloodstains, swinging detachment bloodstains, falling connected bloodstains, and stationary detachment bloodstains (linear); classifying the non-linear scatter bloodstains into non-falling bloodstains and falling bloodstains according to whether there is a radial distribution, classifying the non-falling bloodstains into exhaled bloodstains and non-exhaled bloodstains according to the presence or absence of small rings and bubbles, and classifying the non-exhaled bloodstains into impact scatter bloodstains and stationary detachment bloodstains (non-linear) according to the presence or absence of circular and elliptical bloodstains. Classification may include a second-third classification in which the irregular boundary bloodstains are classified into rubbing bloodstains and rubbing bloodstains depending on whether there are scattering bloodstains or protrusions, the rubbing bloodstains are classified into accumulated bloodstains, massive ejection bloodstains, and massive outflow bloodstains, the rubbing bloodstains are classified into wiped bloodstains and buried bloodstains depending on whether there is a pre-existing bloodstain, and the regular boundary bloodstains are classified into morphological transition bloodstains, absorption bloodstains, flow bloodstains, pooled bloodstains, and irregular transition bloodstains, and the linear ejection bloodstains are classified into wrist linear ejection bloodstains and carotid artery linear ejection bloodstains, the impact scattering bloodstains are classified into blunt force impact scattering bloodstains and gunshot impact scattering bloodstains, and the stop-and-fall bloodstains (non-linear) are classified into anterior stop-and-fall bloodstains and posterior stop-and-fall bloodstains.

[0017] A bloodstain shape classification device using deep learning according to one embodiment of the present invention includes at least one memory for storing at least one instruction and at least one processor, wherein the at least one processor acquires a plurality of bloodstain images by executing the at least one instruction, classifies and labels the acquired bloodstain images based on a predefined classification system, stores the classified and labeled images to create a bloodstain image database, extracts training data and validation data from the bloodstain image database, separates them according to a pre-set ratio, performs training of a deep learning model using the training data, performs validation of the trained deep learning model using the validation data, and performs shape classification of the bloodstain images using the trained deep learning model, wherein the predefined classification system may include a classification system based on the generation mechanism of bloodstains or a morphological classification system of bloodstains.

[0018] According to one embodiment of the present invention, the at least one processor can output a shape classification result of the bloodstain image by executing the at least one instruction.

[0019] According to one embodiment of the present invention, the at least one processor can perform additional training of the trained deep learning model using additional bloodstain images based on the verification result of the trained deep learning model by executing the at least one instruction.

[0020] A computer program stored in a computer-readable storage medium according to one embodiment of the present invention can execute a bloodstain shape classification method using deep learning by using a computer.

[0021] Other aspects, features, and advantages other than those described above will become clear from the following drawings, claims, and description of the invention. Effects of the invention

[0022] According to an embodiment of the present invention, classification can be easily performed when observing bloodstains at the scene, which can contribute to reducing the fatigue of experts during on-site analysis, increasing the speed of analysis, and improving accuracy.

[0023] According to an embodiment of the present invention, by performing classification based on the bloodstain generation mechanism and classification based on the morphological classification system, bloodstain shape classification can be performed in a multifaceted manner. Brief explanation of the drawing

[0024] FIG. 1 is a schematic diagram illustrating a bloodstain shape classification device using deep learning according to one embodiment of the present invention. Figure 2 is a diagram illustrating a classification system based on the mechanism of bloodstain formation. Figure 3 is a diagram illustrating the morphological classification system of bloodstains. FIG. 4 is a flowchart illustrating a bloodstain shape classification method using deep learning according to an embodiment of the present invention. FIG. 5 is a drawing illustrating the output of a display according to one embodiment of the present invention. Specific details for implementing the invention

[0025] Hereinafter, embodiments of the present invention will be described in detail with reference to the attached drawings. When describing with reference to the drawings, identical or corresponding components are given the same reference numerals, and redundant descriptions thereof will be omitted.

[0026] In the following embodiments, terms such as first, second, etc. are used not in a limiting sense, but for the purpose of distinguishing one component from another component.

[0027] In the following examples, singular expressions include plural expressions unless the context clearly indicates otherwise.

[0028] In the following embodiments, terms such as 'include' or 'have' mean that the features or components described in the specification are present, and do not preclude the possibility that one or more other features or components may be added.

[0029] FIG. 1 is a schematic diagram illustrating a bloodstain shape classification device using deep learning according to one embodiment of the present invention.

[0030] Referring to FIG. 1, the bloodstain shape classification device (10) may include a communication unit (110), a processor (120), a memory (130), a user interface (140), and a display (150). Here, the bloodstain shape classification device (10) to which the present invention is applied may be an information processing device used by a user. However, the present invention is not limited thereto, and the bloodstain shape classification device (10) may include additional other components, and some components may be omitted.

[0031] The bloodstain shape classification device (10) may include at least one of a PC (personal computer), a laptop computer, a mobile phone, a tablet PC, a smartphone, a PDA (personal digital assistants), and a PMP (portable multimedia player).

[0032] The communication unit (110) is connected to the processor (120) and memory (130) to transmit and receive data. The communication unit (110) may be connected to other external devices to transmit and receive data. In the following, the expression "transmit and receive A" may indicate transmitting and receiving "information or data representing A".

[0033] The communication unit (110) may be implemented as a circuitry within the bloodstain shape classification device (10). For example, the communication unit (110) may include an internal bus and an external bus. As another example, the communication unit (110) may be an element connecting the bloodstain shape classification device (10) and an external device. The communication unit (110) may be an interface. The communication unit (110) may receive data from an external device and transmit the data to the processor (120) and memory (130). For example, the communication unit (110) may receive a bloodstain image from a shooting device.

[0034] The processor (120) can control the overall operation of the bloodstain shape classification device (10) according to one embodiment of the present invention and perform logical operations.

[0035] The processor (120) processes data received by the communication unit (110) and data stored in memory (130). The processor (120) may be a data processing device implemented in hardware having a circuit having a physical structure for executing desired operations. For example, the desired operations may include code or instructions included in a program.

[0036] The processor (120) executes computer-readable code (e.g., software) stored in memory (e.g., memory (130)) and instructions triggered by the processor (120).

[0037] The processor (120) executes at least one instruction stored in memory (130) to control the performance of a desired operation. Here, the at least one instruction may be stored in internal memory included within the processor (120) or in memory (130) included in a data processing device separate from the processor (120).

[0038] For example, the processor (120) can perform the following operations 210 to 280, and specific details will be described later.

[0039] The memory (130) stores data received by the communication unit (110) and data processed by the processor (120). The memory (130) can store a program (or application, software) that operates the bloodstain shape classification device (10). The stored program can be coded to control the bloodstain shape classification device (10) and can be executed by the processor (120).

[0040] The memory (130) may include volatile memory such as SRAM (Static RAM), DRAM (Dynamic RAM), SDRAM (Synchronous DRAM), or non-volatile memory such as flash memory, PRAM (Phase-change RAM), MRAM (Magnetic RAM), RRAM (Resistive RAM), FRAM (Ferroelectric RAM).

[0041] The user interface (140) can receive user input for controlling the bloodstain shape classification device (10). The user interface (140) may include, but is not limited to, a touch panel that detects a user's touch, a button that receives a user's push operation, a mouse or keyboard for designating or selecting a point on the user interface screen.

[0042] The user interface (140) may be at least one GUI (Graphical User Interface) provided to control, for example, the bloodstain shape classification device (10).

[0043] The display (150) can display information processed by the bloodstain shape classification device (10). For example, the display (150) can display a GUI that controls the bloodstain shape classification device (10).

[0044] The display (150) can be implemented as an LCD, PDP, OLED, FED, LED, flexible display, and 3D display, but is not limited thereto. Additionally, the display (150) can be configured as a touch screen and used as an input device in addition to an output device.

[0045] Additionally, in other embodiments, the bloodstain shape classification device (10) may include more components than those of FIG. 1. For example, the bloodstain shape classification device (10) may include additional components such as a battery and charging device that power the internal components, a database, etc.

[0046] FIG. 2 is a diagram illustrating a classification system based on the generation mechanism of bloodstains. FIG. 3 is a diagram illustrating a morphological classification system of bloodstains. FIG. 4 is a flowchart illustrating a bloodstain shape classification method using deep learning according to an embodiment of the present invention.

[0047] Referring to FIGS. 2 to 4, a bloodstain shape classification method (20) using deep learning according to one embodiment of the present invention will be described.

[0048] The following operations can be performed by the aforementioned bloodstain shape classification device (10).

[0049] In operation 210, the bloodstain shape classification device (10) can acquire multiple bloodstain images.

[0050] The bloodstain shape classification device (10) can receive a plurality of bloodstain images obtained from a shooting device through a communication unit (110). Additionally, the bloodstain shape classification device (10) may include a camera including a lens and an image sensor. The image sensor can convert an image input by the lens into an electrical signal and may be a semiconductor device such as a CCD or CMOS.

[0051] Multiple bloodstain images may include original bloodstain images and extended bloodstain images.

[0052] The acquired bloodstain images may include real-life images taken with a camera, scanned document images, etc.

[0053] The acquired bloodstain images undergo a preprocessing process. This preprocessing may include resizing, noise removal, brightness / contrast normalization, and rotation / inversion correction. This is intended to improve the accuracy of subsequent classification tasks and ensure data consistency.

[0054] In addition, for more accurate classification, an extended bloodstain image can be generated for each original bloodstain image. An extended bloodstain image can be generated through various transformations and processing based on the acquired original bloodstain image. The extended bloodstain image can be generated by methods such as rotation, resizing, color conversion, and adding noise to the original bloodstain image.

[0055] According to one embodiment of the present invention, an expanded bloodstain image can be generated by dividing an original bloodstain image into a plurality of regions and applying an opaque filter to the divided bloodstain images.

[0056] For example, an expanded bloodstain image can be generated by applying a global opacity filter, a one-way gradient opacity filter, a two-way gradient opacity filter, etc., to a segmented bloodstain image. In addition, the segmented regions or the segmented regions can be combined and stored in a database.

[0057] In operation 220, the acquired bloodstain images can be classified and labeled based on a predefined classification system.

[0058] Here, the predefined classification system may be a classification system based on the mechanism of bloodstain formation or a morphological classification system.

[0059] Referring to Fig. 2, the classification system based on the mechanism of bloodstain formation may include the following 1-1, 1-2, 1-3, and 1-4 classifications.

[0060] Classification 1-1 can classify bloodstains into spattered bloodstains and non-spattered bloodstains.

[0061] Category 1-2 is a subcategory of Category 1-1 and is as follows.

[0062] Dispersed hemorrhages can be classified according to the generation mechanisms of gravitational action, dispersion of external forces, vascular pressure ejection, and escape. Non-dispersed hemorrhages can be classified according to the generation mechanisms of surface action and transference.

[0063] Category 1-3 is a subcategory of Category 1-2 and is as follows.

[0064] Bloodstains caused by the action of gravity can be classified into drop bloodstains, drop-connected bloodstains, and accumulated bloodstains. Bloodstains caused by the dispersion of external force can be classified into impact scatter bloodstains and exhalation bloodstains. Bloodstains caused by the ejection of vascular pressure can be classified into linear ejection bloodstains, massive ejection bloodstains, massive outflow bloodstains, and arterial non-bloodstains. Bloodstains caused by detachment can be classified as detachment bloodstains.

[0065] In addition, bloodstains caused by surface action can be classified into pooled bloodstains, flow bloodstains, and absorption bloodstains. Bloodstains caused by transfer can be classified as contact bloodstains.

[0066] Category 1-4 is a subcategory of Category 1-3 and is as follows.

[0067] Impact spatter bloodstains can be classified into blunt force gunshot spatter bloodstains and firearm impact spatter bloodstains. Linear spurting bloodstains can be classified into wrist linear spurting bloodstains and carotid artery linear spurting bloodstains. Disengaged bloodstains can be classified into swinging disengaged bloodstains and stationary disengaged bloodstains. Here, stationary disengaged bloodstains can be classified into anterior stationary disengaged bloodstains and posterior stationary disengaged bloodstains. Contact bloodstains can be classified into shape-transition bloodstains, irregular shape-transition bloodstains, and rubbing bloodstains. Here, rubbing bloodstains can be classified into wiped bloodstains and smeared bloodstains.

[0068] Referring to FIG. 3, the morphological classification system of bloodstains may include the following 2-1, 2-2, 2-3, and 2-4 classifications.

[0069] Classification 2-1 can classify bloodstains into scattered bloodstains and non-scattered bloodstains according to the morphological characteristics of the bloodstains.

[0070] Category 2-2 is a subcategory of Category 2-1 and is as follows.

[0071] Bloodsplatter can be classified into linear bloodsplatter and non-linear bloodsplatter depending on whether there is a linear distribution. Non-bloodsplatter can be classified into irregular boundary bloodsplatter and regular boundary bloodsplatter based on the regularity of the boundaries.

[0072] Category 2-3 is a subcategory of Category 2-2 and is as follows.

[0073] Linear spatter bloodstains can be classified into linear ejection bloodstains, swinging detachment bloodstains, falling connected bloodstains, and stationary detachment bloodstains (linear).

[0074] Non-linear bloodstains can be classified into non-falling bloodstains and falling bloodstains depending on the presence or absence of a radial distribution. Here, non-falling bloodstains can be classified into exhaled bloodstains and non-exhaled bloodstains depending on the presence or absence of small rings and bubbles. Non-exhaled bloodstains can be classified into impact bloodstains and stationary displaced bloodstains (non-linear) depending on the presence or absence of circular and elliptical bloodstains.

[0075] Irregular boundary bloodstains can be classified into rubbing bloodstains and rubbing bloodstains depending on the presence or absence of spatters or protrusions. Rubbing bloodstains can be classified into accumulated bloodstains, massive spurting bloodstains, and massive effusion bloodstains. Rubbing bloodstains can be classified into wiped bloodstains and buried bloodstains depending on the presence or absence of pre-existing bloodstains.

[0076] Regular boundary bloodstains can be classified into morphological transfer bloodstains, absorption bloodstains, flow bloodstains, pooled bloodstains, and irregular transfer bloodstains.

[0077] Category 2-4 is a subcategory of Category 2-3 and is as follows.

[0078] Linear spurt hemorrhages can be classified into wrist linear spurts and carotid artery linear spurts. Impact scatter hemorrhages can be classified into blunt force impact scatter hemorrhages and firearm impact scatter hemorrhages. Stoppage hemorrhages (non-linear) can be classified into anterior stoppage hemorrhages and posterior stoppage hemorrhages.

[0079] Meanwhile, types of unclassified bloodstains not classified according to the above classification systems include capillary bloodstains, coagulated bloodstains, contour bloodstains, space marks, insect bloodstains, secondary splatter bloodstains, and deformed bloodstains.

[0080] As shown above, labeling can be performed on bloodstain images classified by each classification system.

[0081] Labeling is the process of identifying the content and features of an image and tagging them in a form suitable for machine learning.

[0082] Operation 230 can create a bloodstain image database by storing classified and labeled bloodstain images.

[0083] Classified and labeled bloodstain images can be stored in a hierarchical structure according to the assigned classification system and labels. Each bloodstain image is assigned a unique identifier, and metadata may be recorded along with it. For example, the metadata may include classification information and labels of the bloodstain image, information distinguishing between the original and expanded bloodstain images, preprocessing history, transformation information used for data expansion, and characteristic information of the bloodstain image (size, format, etc.). Metadata facilitates searching, filtering, and management of the database, and enables efficient data selection and utilization during the training process of deep learning models.

[0084] In addition, the bloodstain image database can be designed with an expandable structure to facilitate the addition of new bloodstain images or classification systems.

[0085] Operation 240 can extract training data and validation data from a bloodstain image database and separate them according to a preset ratio.

[0086] Training a deep learning model requires a sufficient amount of data, which is divided into training data and validation data. Training data is used for the model to learn patterns, while validation data is used to evaluate the performance of the trained model.

[0087] Extract data suitable for training from the bloodstain image database.

[0088] Afterwards, the data is separated into training data and validation data according to a pre-set ratio (e.g., 80:20 or 70:30, etc.).

[0089] This process can be performed automatically without manual intervention and can be applied based on random sampling of data or specific criteria (e.g., maintaining class balance). These data splitting methods can contribute to improving the accuracy of learning and preventing overfitting.

[0090] In operation 250, deep learning model training can be performed using training data.

[0091] Deep learning is a field of machine learning that utilizes multilayer neural networks to learn data and automatically extract features to recognize patterns. For example, deep learning models applicable to the embodiments of the present invention include CNN (Convolutional Neural Network) and ResNet (Residual Network).

[0092] Training data can be input into a deep learning model, and weights can be optimized using a loss function and an optimizer.

[0093] The loss function can quantify and measure the difference (loss) between the actual and predicted values ​​of a deep learning model. The weights and biases of a deep learning model are updated based on the loss function, and optimizers (e.g., Adam, SGD) can optimize the model using gradient descent.

[0094] Optimal performance can be achieved by adjusting hyperparameters such as the learning rate, batch size, and number of epochs during the training process of a deep learning model.

[0095] In operation 260, verification of the trained deep learning model can be performed using the verification data.

[0096] The performance of the model can be evaluated using validation data that was not used as training data. During the validation process, evaluation metrics can be calculated by comparing the predicted values ​​output by the deep learning model with the actual correct values. These evaluation metrics may include accuracy, precision, recall, F1-score, and loss function values.

[0097] For example, the deep learning model uses ResNet-101, and training is performed over 100,000 epochs. The learning rate is set to 0.001, and validation can be performed in 10 cycles.

[0098] After the operation (260) of performing verification of the trained deep learning model, the next operation 265 may be added.

[0099] In operation 265, based on the validation results of the trained deep learning model, additional training of the trained deep learning model can be performed using additional bloodstain images.

[0100] Here, additional bloodstain images are bloodstain images that were not used for training and verification.

[0101] In the additional training process, the performance of the model can be optimized by retraining the existing deep learning model using additional bloodstain images. To this end, the additional bloodstain images can be used as training data after undergoing a preprocessing process to improve data quality.

[0102] In addition, according to one embodiment of the present invention, instead of retraining all weights of an existing deep learning model, a fine-tuning method can be applied to adjust only the weights for a specific layer of the deep learning model.

[0103] In operation 270, shape classification of bloodstain images can be performed using a trained deep learning model.

[0104] Using a trained deep learning model, key features can be extracted from an input bloodstain image, and based on the key features, shape classification of the bloodstain image can be performed.

[0105] The classification of the shape of bloodstain images may include a classification system based on the bloodstain generation mechanism described above or a classification system based on the morphological classification of bloodstains.

[0106] In operation 280, the bloodstain shape classification device (10) can output a classification result.

[0107] The operation (280) of outputting the classification result may include an operation of displaying the shape classification result of the bloodstain image (e.g., classification by bloodstain generation mechanism, morphological classification of bloodstain) respectively. For example, the classification results based on the bloodstain generation mechanism classification system and the bloodstain morphological classification system may be output step by step.

[0108] FIG. 5 is a drawing illustrating the output of a display according to one embodiment of the present invention.

[0109] For example, referring to FIG. 5, the shape classification results for a specific bloodstain image can be displayed on the display (150). When a user selects a bloodstain image through the 'Image' button (151) and classifies it through the 'Classification' button (153), it can be displayed as 'Classification system by generation mechanism: spatter bloodstain - vascular pressure ejection - linear ejection bloodstain - wrist linear ejection bloodstain', 'Morphological classification system: spatter bloodstain - linear spatter bloodstain - linear ejection bloodstain - wrist linear ejection bloodstain'.

[0110] As another example, the GUI may be used to highlight nodes corresponding to specific classification results on the logic tree. Node highlighting can be achieved by changing the node's color or modifying its configuration (node ​​shape, font weight, etc.). Highlighting can also be achieved by expanding or contracting the node's size. Thus, the user can check the location of the corresponding classification result on the logic tree.

[0111] Meanwhile, the bloodstain shape classification method (20) using deep learning according to various embodiments of the present invention illustrated in FIG. 4 can be written as a program that can be executed on a computer and can be implemented on a general-purpose digital computer that operates the program using a computer-readable recording medium. The computer-readable recording medium may include storage media such as magnetic storage media (e.g., ROM, floppy disk, hard disk, etc.) and optical reading media (e.g., CD-ROM, DVD, etc.).

[0112] The above description is merely an illustrative explanation of the technical concept of the present invention, and those skilled in the art to which the present invention pertains will be able to make various modifications and variations within the scope of the essential characteristics of the present invention. For example, appropriate results may be achieved even if the described techniques are performed in a different order than the described method, and / or the components of the described system, structure, device, circuit, etc. are combined or assembled in a form different from the described method, or are replaced or substituted by other components or equivalents.

[0113] Accordingly, the scope of protection of the present invention shall be interpreted by the claims below, and all technical ideas within an equivalent scope shall be interpreted as being included within the scope of rights of the present invention. Explanation of the symbols

[0114] 10: Bloodstain shape classification device 110 : Communications Department 120 : Processor 130 : Memory 140 : User Interface 150 : Display

Claims

Claim 1 A method for classifying bloodstain shapes using deep learning, comprising: acquiring a plurality of bloodstain images; classifying and labeling the acquired bloodstain images based on a predefined classification system; creating a bloodstain image database by storing the classified and labeled images; extracting training data and validation data from the bloodstain image database and separating them according to a preset ratio; training a deep learning model using the training data; validating the trained deep learning model using the validation data; and classifying the shapes of the bloodstain images using the trained deep learning model; wherein the predefined classification system includes a classification system based on the generation mechanism of the bloodstain or a morphological classification system of the bloodstain. Claim 2 A method for classifying bloodstain shapes using deep learning, wherein the bloodstain image includes an original bloodstain image and an extended bloodstain image. Claim 3 A method for classifying bloodstain shapes using deep learning, wherein the expanded bloodstain image is generated by dividing the original bloodstain image into multiple regions and applying an opaque filter to the divided bloodstain images. Claim 4 A bloodstain shape classification method using deep learning, further comprising, in claim 1, the operation of outputting a shape classification result of the bloodstain image. Claim 5 A method for classifying bloodstain shapes using deep learning, comprising: the operation of performing additional training of the trained deep learning model using additional bloodstain images based on the verification result of the trained deep learning model in claim 1. Claim 6 In claim 1, the classification system based on the generation mechanism of the bloodstain comprises: Classification 1-1, which classifies bloodstains into scattering bloodstains and non-scattering bloodstains; Classification 1-2, which classifies the scattering bloodstains according to the generation mechanisms of gravitational action, external force dispersion, vascular pressure ejection, and detachment, and classifies the non-scattering bloodstains according to the generation mechanisms of surface action and transference; Bloodstains caused by gravitational action are classified into falling bloodstains, falling connected bloodstains, and accumulated bloodstains; Bloodstains caused by external force dispersion are classified into impact scattering bloodstains and exhalation bloodstains; Bloodstains caused by vascular pressure ejection are classified into linear ejection bloodstains, massive ejection bloodstains, massive outflow bloodstains, and arterial non-bloodstains; Bloodstains caused by detachment are classified as detachment bloodstains; Bloodstains caused by surface action are classified into pooled bloodstains, flow bloodstains, and absorption bloodstains; and Bloodstains caused by transference A method for classifying bloodstain shapes using deep learning, comprising: a first-3 classification classifying bloodstains as contact bloodstains; and a first-4 classification classifying impact scattering bloodstains into blunt force gunshot scattering bloodstains and gunshot impact scattering bloodstains, classifying linear spurting bloodstains into wrist linear spurting bloodstains and carotid artery linear spurting bloodstains, classifying detaching bloodstains into swinging detaching bloodstains and stopping detaching bloodstains, classifying stopping detaching bloodstains into forward stopping detaching bloodstains and rear stopping detaching bloodstains, classifying contact bloodstains into shape transition bloodstains, irregular shape transition bloodstains, and rubbing bloodstains, and classifying rubbing bloodstains into wiped bloodstains and buried bloodstains. Claim 7 In paragraph 1, the morphological classification system of the bloodstain comprises: a 2-1 classification that classifies the bloodstain into scattering bloodstains and non-scattering bloodstains according to the morphological characteristics of the bloodstain; a 2-2 classification that classifies the scattering bloodstains into linear scattering bloodstains and non-linear scattering bloodstains according to whether there is a linear distribution, and classifies the non-scattering bloodstains into irregular boundary bloodstains and regular boundary bloodstains according to whether there is regularity of the boundary; The above linear scatter bloodstains are classified into linear spurting bloodstains, swinging detachment bloodstains, falling connected bloodstains, and stationary detachment bloodstains (linear); the above non-linear scatter bloodstains are classified into non-falling bloodstains and falling bloodstains according to the presence or absence of radial distribution; the above non-falling bloodstains are classified into exhaled bloodstains and non-exhaled bloodstains according to the presence or absence of small rings and bubbles; non-exhaled bloodstains are classified into impact scatter bloodstains and stationary detachment bloodstains (non-linear) according to the presence or absence of circular and elliptical bloodstains; the above irregular boundary bloodstains are classified into rubbing bloodstains and rubbing bloodstains according to the presence or absence of scatter bloodstains or protrusions; the above rubbing bloodstains are classified into accumulated bloodstains, massive spurting bloodstains, and massive outflow bloodstains; the above rubbing bloodstains are classified into wiped bloodstains and buried bloodstains according to the presence or absence of pre-existing bloodstains; and the above regular boundary bloodstains A method for classifying bloodstain shapes using deep learning, comprising: a second-third classification classifying into shape-transition bloodstains, absorption bloodstains, flow bloodstains, pooled bloodstains, and irregular shape-transition bloodstains; and a second-fourth classification classifying the linear ejection bloodstains into wrist linear ejection bloodstains and carotid artery linear ejection bloodstains, classifying the impact scattering bloodstains into blunt force impact scattering bloodstains and gunshot impact scattering bloodstains, and classifying the stop-and-go bloodstains (non-linear) into anterior stop-and-go bloodstains and posterior stop-and-go bloodstains. Claim 8 A bloodstain shape classification device using deep learning, comprising: at least one memory storing at least one instruction; and at least one processor; wherein the at least one processor, by executing the at least one instruction, acquires a plurality of bloodstain images, classifies and labels the acquired bloodstain images based on a predefined classification system, stores the classified and labeled images to create a bloodstain image database, extracts training data and validation data from the bloodstain image database, separates them according to a preset ratio, performs training of a deep learning model using the training data, performs validation of the trained deep learning model using the validation data, and performs shape classification of the bloodstain images using the trained deep learning model; wherein the predefined classification system includes a classification system based on the bloodstain generation mechanism or a morphological classification system of bloodstains. Claim 9 A bloodstain shape classification device using deep learning, wherein the bloodstain image includes an original bloodstain image and an extended bloodstain image. Claim 10 A bloodstain shape classification device using deep learning, wherein the expanded bloodstain image is generated by dividing the original bloodstain image into a plurality of regions and applying an opaque filter to the divided bloodstain images. Claim 11 A bloodstain shape classification device using deep learning, wherein, in claim 8, the at least one processor outputs a shape classification result of the bloodstain image by executing the at least one instruction. Claim 12 A bloodstain shape classification device using deep learning, wherein, in claim 8, the at least one processor performs additional training of the trained deep learning model using additional bloodstain images based on the verification result of the trained deep learning model by executing the at least one instruction. Claim 13 In paragraph 8, the classification system based on the generation mechanism of the bloodstain comprises: Classification 1-1, which classifies bloodstains into scattering bloodstains and non-scattering bloodstains; Classification 1-2, which classifies the scattering bloodstains according to the generation mechanisms of gravitational action, external force dispersion, vascular pressure ejection, and detachment, and classifies the non-scattering bloodstains according to the generation mechanisms of surface action and transference; Bloodstains caused by gravitational action are classified into falling bloodstains, falling connected bloodstains, and accumulated bloodstains; Bloodstains caused by external force dispersion are classified into impact scattering bloodstains and exhalation bloodstains; Bloodstains caused by vascular pressure ejection are classified into linear ejection bloodstains, massive ejection bloodstains, massive outflow bloodstains, and arterial non-bloodstains; Bloodstains caused by detachment are classified as detachment bloodstains; Bloodstains caused by surface action are classified into pooled bloodstains, flow bloodstains, and absorption bloodstains; and Bloodstains caused by transference A bloodstain shape classification device using deep learning, comprising: a first-3 classification classifying bloodstains as contact bloodstains; and a first-4 classification classifying impact scattering bloodstains into blunt force gunshot scattering bloodstains and gunshot impact scattering bloodstains, linear spurting bloodstains into wrist linear spurting bloodstains and carotid artery linear spurting bloodstains, detaching bloodstains into swinging detaching bloodstains and stopping detaching bloodstains, stopping detaching bloodstains into forward stopping detaching bloodstains and rear stopping detaching bloodstains, contact bloodstains into shape transition bloodstains, irregular shape transition bloodstains, and rubbing bloodstains, and rubbing bloodstains into wiped bloodstains and buried bloodstains. Claim 14 In paragraph 8, the morphological classification system of the bloodstain comprises: Classification 2-1, which classifies the bloodstain into scattering bloodstains and non-scattering bloodstains according to the morphological characteristics of the bloodstain; Classification 2-2, which classifies the scattering bloodstains into linear scattering bloodstains and non-linear scattering bloodstains according to whether there is a linear distribution, and classifies the non-scattering bloodstains into irregular boundary bloodstains and regular boundary bloodstains according to whether there is regularity of the boundary; The above linear scatter bloodstains are classified into linear spurting bloodstains, swinging detachment bloodstains, falling connected bloodstains, and stationary detachment bloodstains (linear); the above non-linear scatter bloodstains are classified into non-falling bloodstains and falling bloodstains according to the presence or absence of radial distribution; the above non-falling bloodstains are classified into exhaled bloodstains and non-exhaled bloodstains according to the presence or absence of small rings and bubbles; non-exhaled bloodstains are classified into impact scatter bloodstains and stationary detachment bloodstains (non-linear) according to the presence or absence of circular and elliptical bloodstains; the above irregular boundary bloodstains are classified into rubbing bloodstains and rubbing bloodstains according to the presence or absence of scatter bloodstains or protrusions; the above rubbing bloodstains are classified into accumulated bloodstains, massive spurting bloodstains, and massive outflow bloodstains; the above rubbing bloodstains are classified into wiped bloodstains and buried bloodstains according to the presence or absence of pre-existing bloodstains; and the above regular boundary bloodstains A bloodstain shape classification device using deep learning, comprising: a second-third classification classifying into shape-transition bloodstains, absorption bloodstains, flow bloodstains, pooled bloodstains, and irregular shape-transition bloodstains; and a second-fourth classification classifying the linear ejection bloodstains into wrist linear ejection bloodstains and carotid artery linear ejection bloodstains, classifying the impact scattering bloodstains into blunt force impact scattering bloodstains and gunshot impact scattering bloodstains, and classifying the stop-and-go bloodstains (non-linear) into anterior stop-and-go bloodstains and posterior stop-and-go bloodstains. Claim 15 A computer program stored on a computer-readable storage medium to execute the method of any one of claims 1 through 7 using a computer.

Images