System for insect mortality assessment and related methods

Abstract

A system for insect mortality evaluation is provided. In one example, the system includes one or more cameras and a removable plate positioned at a predetermined distance from the cameras. The removable plate has a hole in which an insect specimen is disposed. A memory stores instructions that, when executed by a processor, cause the processor to transmit one or more signals to the one or more cameras to capture a plurality of time series images of the insect specimen and pre-process the time series images of the insect specimen. The instructions further cause the system to analyze the time series images of the insect specimen using a trained Siamese neural network to determine a state of viability of the insect specimen and store a result indicative of the state of viability of the insect specimen in the memory.

Description

Cross-references to related applications

[0001] This application claims the benefit of U.S. Provisional Patent Application No. 63 / 556,969, filed February 23, 2024, the disclosure of which is hereby incorporated by reference in its entirety. Technical Field

[0002] This disclosure generally relates to systems and methods for automatically distinguishing between live and non-live insect specimens. Systems and methods for assessing insect mortality rates are also disclosed. Background Technology

[0003] This section provides background information relating to this disclosure, which is not necessarily prior art.

[0004] In agriculture, and more specifically in crop protection, it is known to test the efficacy of pesticides against target pests. For example, during pesticide discovery or development, it may be necessary to test the insecticidal properties of multiple compounds of interest against target insects in a controlled / laboratory setting. Such assessments of insecticidal properties typically involve evaluating the mortality rate of exposed insects, which in turn requires distinguishing between live and non-live / dead insect specimens. Distinguishing between live and dead insect specimens is usually image-based, but can be time-consuming and error-prone. Image-based automated mortality assessments can be performed, which may be more efficient and accurate, and thus allow pesticide discovery to proceed at a faster pace. Summary of the Invention

[0005] This section provides a general overview of this disclosure and is not a full disclosure of its entire scope or all its features.

[0006] The exemplary embodiments of this disclosure generally relate to a system for detecting insect mortality. In one exemplary embodiment, such a system typically includes one or more cameras and a removable plate located at a predetermined distance from the one or more cameras. The removable plate has holes in which insect specimens are disposed. The exemplary system also includes a processor and a memory storing instructions that, when executed by the processor, cause the processor to: transmit one or more signals to the one or more cameras to capture at least two time-series images of the insect specimen; and preprocess the at least two time-series images of the insect specimen. The instructions further cause the system to: analyze the at least two time-series images of the insect specimen using a trained Siamese neural network to determine the viability state of the insect specimen; and store the results indicating the viability state of the insect specimen in the memory.

[0007] The exemplary embodiments of this disclosure also generally relate to methods for detecting insect mortality. In one exemplary embodiment, the method typically includes removing a removable plate located at a predetermined distance from one or more cameras, the removable plate having holes in which insect specimens are disposed. The exemplary method further includes: transmitting one or more signals to the one or more cameras to capture at least two time-series images of the insect specimen; and preprocessing the at least two time-series images of the insect specimen. The exemplary method further includes: analyzing the at least two time-series images of the insect specimen using a trained Siamese neural network to determine the viability state of the insect specimen; and storing the result indicating the viability state of the insect specimen in a memory coupled to a processor.

[0008] Further applicability will become apparent from the description provided herein. The descriptions and specific examples within the scope of this invention are intended for illustrative purposes only and are not intended to limit the scope of this disclosure. Attached Figure Description

[0009] The accompanying drawings described herein are for illustrative purposes only, representing selected embodiments and not all possible implementations, and are not intended to limit the scope of this disclosure.

[0010] Figure 1 This is an example implementation of a system for detecting insect mortality rates; Figure 2 Example time-series images of insect specimens are shown; Figure 3 An example method for detecting insect mortality rates according to existing technology is shown; Figure 4 This is an example implementation of a method for detecting insect mortality rates; Figure 5 The dataset used in an example implementation of the system disclosed herein is shown; Figure 6 Data generated using the system disclosed herein is shown, and the accuracy of the system is demonstrated. Figure 7 A confusion matrix associated with data generated using the system of this disclosure is shown; Figure 8 Examples of what can be combined with Figure 1 An example configuration of computing devices used together with the system shown; Figure 9 This is a flowchart of an example process for detecting insect mortality. Detailed Implementation

[0011] Example embodiments will now be described more fully with reference to the accompanying drawings. The descriptions and specific examples included herein are intended for illustrative purposes only and are not intended to limit the scope of this disclosure.

[0012] Figure 1 An example system 100 for distinguishing between live and non-live insect specimens is shown. The example system 100 typically includes a well plate 101, which in this example embodiment is a 12-well plate. In other embodiments, the well plate 101 may be a 6-well plate, a 24-well plate, or may include any number of wells or containers that enable the example system 100 to function as described herein. The well plate 101 includes occupied wells 102 containing insect specimens 104 surrounded by a liquid medium. The liquid medium may include a candidate insecticide, the insecticidal efficacy of which is being tested against the insect specimens 104. The liquid medium may include other components such as water, nutrients, pH buffers, etc. In connection with certain control and / or comparative experiments, the liquid medium may not include an insecticide. In other embodiments, as an alternative to or supplement to the liquid medium, the occupied wells 102 may contain a gel.

[0013] Example system 100 also includes a camera 103 configured to collect images of the perforated plate 101. Images of the perforated plate 101 may include substantially all of the perforated plate 101, or may include portions of the perforated plate 101 (e.g., individual holes of the perforated plate 101). In the example embodiment, camera 103 is a digital camera. In some embodiments, system 100 includes multiple cameras 103. In some embodiments, camera 103 is operatively coupled to a computing device (e.g., a computing device). Figure 8 The computing device 810 shown or combined with the computing device.

[0014] Figure 1 It also shows the use of camera 103 in time An example first image 105 is taken of the occupied aperture 102. Within the first image 105 is an image of the insect specimen 104. In some embodiments, the system 100 may include a memory for storing digital images collected from the camera 103. The stored digital images can be analyzed for, for example... The size, shape and / or location of the insect specimen 104 within the occupied hole 102 are characterized. Figure 1 It was also shown at a later time An example second image 106 of the occupied aperture 102 taken at the location. Similar to the first image 105, the second image 106 can be stored in memory and analyzed for... The insect specimens 104 inside the occupied holes 102 are characterized.

[0015] Generally, a comparison between the first image 105 and the second image 106 can provide information about the viability of the insect specimen 104. For example, the position of the insect specimen 104 in each of the example images 105 and 106 can indicate a loss of activity, which can further indicate that the insect specimen 104 is dead. This mortality assessment can be used to determine the insecticidal efficacy of the pesticide present in the liquid medium in which the insect specimen 104 is exposed. In this example, it can be concluded that the insecticidal efficacy of the pesticide present in the liquid medium is sufficient for the insect specimen 104.

[0016] Figure 2 Another example first image 205 and an example second image 206 are shown. First image 205 was collected after insect specimen 204 was placed in a hole 202 filled with a liquid medium. In this example, the liquid medium in hole 202 includes an insecticide. Insect specimen 204 was allowed to remain in the hole for 3 days before collecting first image 205. Second image 206 of hole 202 was collected one hour after collecting first image 205. The comparison between first image 205 and second image 206 can be used to assess the mortality rate of insect specimen 204. For example, it can be observed that between collecting first image 205 and collecting second image 206, insect specimen 204 appears to have moved from the left half of hole 202 to the right half of hole 202. This apparent activity indicates the viability of insect specimen 204 and leads the observer to conclude that insect specimen 204 is alive. Compared to the previous... Figure 1 As described above, this mortality assessment can be used to determine the insecticidal efficacy of the pesticide present in the liquid medium to which insect specimen 204 is exposed. In this example, it can be concluded that the insecticidal efficacy of the pesticide present in the liquid medium is insufficient for insect specimen 104.

[0017] Figure 3 An example mortality assessment process as disclosed in the prior art is illustrated. A first image 305 is collected after an insect specimen 304 is placed in a hole 302 filled with a liquid medium. The liquid medium in the hole 302 may or may not include an insecticide. The insect specimen 304 is allowed to remain in the hole for 3 days before collecting the first image 305. A second image 306 of the hole 302 is collected one hour after the collection of the first image 305. Images 305 and 306 can be visually compared for mortality assessment. Alternatively, the assessment can be performed automatically using a computer algorithm combined with a trained dataset. For example, an algorithm can be used to subtract images 305 and 306, and a corresponding classifier trained on the trained dataset can be used to assign the insect specimen 304 a "dead" or "surviving" status.

[0018] Further reference Figure 3The algorithm can process images 305 and 306, thereby assigning a numerical value to each pixel in each of images 305 and 306. For example, an algorithm can be trained and used to distinguish pixels associated with insect specimen 304 from pixels associated with the surrounding medium (i.e., not associated with insect specimen 304). This algorithm can be called a masking algorithm. Image 307 is derived from image 305 and generated by, for example, assigning "1" to each pixel associated with insect specimen 304 and assigning "0" to each pixel associated with the surrounding medium. Image 308 is similarly generated using image 306 as input. The algorithm then compares corresponding pixels in images 307 and 308 and assigns further values ​​to each pair of corresponding pixels. For example, the value "0" can be assigned to each pair of corresponding pixels with equal values ​​(i.e., both 0 or 1), and the value "1" can be assigned to each pair of corresponding pixels with unequal values. A sum including all pixel pairs can then be calculated, and the value of this sum can be compared with a threshold and used for mortality assessment. For example, comparing images 307 and 308 as described above will produce a sum greater than zero because there are many pairs of corresponding pixels including pixels assigned different values. If the sum is sufficiently greater than a certain threshold distance from zero, the algorithm can mark insect specimen 304 as alive. Conversely, an insufficient sum may cause the algorithm to mark insect specimen 304 as dead. However, the prior art method described above has numerous drawbacks. For example, such methods are generally not robust to changes in lighting conditions. That is, in one case, the lighting conditions of image 307 may be slightly different from those of image 308 due to, for example, a slight change in the angle between camera 103 and aperture 102. In another case, image 307 may be brighter than image 308 due to external environmental factors, or vice versa. This may cause corresponding pixels in images 307 and 308 to have different values ​​after the masking step, when they should be the same, which may produce false positive results (e.g., the algorithm incorrectly determines that the insect specimen is alive when it is actually dead). A similar problem may arise when aperture 102 is offset by a certain distance between the moments when images 307 and 308 are captured. Therefore, there is a need in the art for a more robust algorithm for determining insect mortality.

[0019] Figure 4 This is an example implementation of a method 400 for detecting insect mortality, and can be used in conjunction with system 100. Figure 8 The computing device 810 shown is used to implement this. The first image 405 is taken after the insect specimen 404 is placed in the hole 402 filled with a liquid medium. The insect specimen 404 is collected at the well 402. The liquid medium in well 402 may or may not include pesticides. The insect specimen 404 is allowed to remain in the well for a period of time (e.g., 3 days) before collecting the first image 405. The second image 406 of well 402 is collected after the first image 405. Collected from [location]. In some implementations, time [is used]. For time The subsequent 0.5, 1.0, 2.0, 3.0, 10, 24, or 48 hours. In some embodiments, the first image 405 and the second image 406 are two images from a set of more than two time-series images. For example, it can be... and Multiple images are collected at each location, and can be used in addition to and Images were collected at times other than those specified.

[0020] In some embodiments, the first image 405 and the second image 406 may be preprocessed. Preprocessing may include removing the outer edges of portions of digital images 405 and 406 that do not contain the insect specimen, or removing irrelevant visual data (i.e., image cropping). In other embodiments, preprocessing may include changing the aspect ratio of digital images 405 and 406, enlarging, separating portions of the images, or any other process that enables the method to be performed as described herein.

[0021] The first image 405 is analyzed by a Siamese Neural Network (SNN) 420. In an example embodiment, the SNN 420 is a trained neural network trained on a first set of live specimen images and a second set of dead specimen images. In some embodiments, the live specimen images are oversampled compared to the dead specimen images. In some embodiments, the dead specimen images are oversampled, and in some embodiments, the images of live and dead specimens are sampled approximately equally. The SNN 420 can be retrained as new or more digital images or data derived therefrom become available.

[0022] In some embodiments, the first set of live specimen images includes one or more original images of the live specimen and an augmented dataset. The augmented dataset can be generated by modifying one or more original images of the live specimen. In one embodiment, the modification may be to offset pixels in one or more original images of the live specimen by a predetermined distance. For example, in each original image, each pixel may be offset to the left by 5 pixels. In another embodiment, the augmented dataset can be generated by modifying the lighting conditions of one or more original images (e.g., increasing the brightness of the images).

[0023] In an example implementation, the SNN 420 includes two convolutional neural networks (CNNs), including CNN 410. The SNN 420 may also be referred to as a Siamese neural network. CNN 410 is configured to analyze digital images, more specifically, to extract features from the digital images to feed into a statistical model. In some implementations, CNN 410 is a feedforward neural network, a fully connected neural network, and may include any number of layers and kernels that enable CNN 410 to function as described herein. CNN 410 may also include any number of neurons and weights for each neuron in a given layer, enabling CNN 410 to function as described herein. In some implementations, CNN 410 is a ResNet50 network.

[0024] The second image 406 is also analyzed by the SNN 420 in a similar manner to the first image 405. More specifically, the second image 406 is processed using a CNN 411 similar to CNN 410. For example, CNN 410 and CNN 411 may use the same weights and may share other common features. In some embodiments, CNN 411 is a ResNet50 network. In some embodiments, CNN 410 and CNN 411 are substantially the same and may be substantially the same ResNet50 network. In this case, for them to be “substantially the same,” CNN 410 and CNN 411 are sufficiently similar to enable the accurate determination of insect specimen mortality as disclosed herein.

[0025] Using a trained SNN 420, embedding 431 is generated for the first image 405, and a separate embedding 432 is generated for the second image 406. Generally, a set of embeddings is generated by the trained SNN 420, each embedding uniquely corresponding to a time-series image of the insect specimen 404. In the example implementation, embedding refers to the process of mapping categorical variables to numbers constituting a vector. Therefore, embeddings can reduce the dimensionality of categorical variables. The embedding dimension can be selected or changed based on the performance of method 400.

[0026] Further reference Figure 4 The difference 440 between embeddings 431 and 432 is calculated. Combined with classifier layer 450, the difference 440 can be used to determine the viability state of insect specimen 404. In some embodiments, classifier layer 450 is a sigmoid classifier layer. In some embodiments, classifier layer 450 is a logistics classifier layer.

[0027] Figure 5 A dataset used in an example implementation of the system of this disclosure is shown. As a utilization... Figure 4 The methods and algorithms shown are part of the experiments, such as Figure 5As shown, SNN 420 was trained on three separate instances across three different datasets. For example, in the first experiment, SNN 420 was trained using the original dataset with 3,200 images of surviving specimens and 640 images of dead specimens. The trained model was validated against 561 images of surviving specimens and 172 images of dead specimens. Additionally, after validation, the trained model was tested on 342 images of surviving specimens and 129 images of dead specimens. In alternative experiments, the surviving specimen images were undersampled and oversampled. In the undersampling experiment, 1,280 images of surviving specimens were used. In the oversampling experiment, 6,400 images of surviving specimens were used. For this particular embodiment of the present disclosure, it has been found that the model trained on the oversampled dataset performs better.

[0028] Figure 6 Data generated using the system of this disclosure and used to demonstrate the accuracy of the system are shown, including data table 601 and a corresponding bar chart 602. As shown in data table 601, the methods of this disclosure (e.g., Figure 4 The method shown was used to analyze the training data. More specifically, mortality assessment was performed on 1,270 training samples. The assessment correctly assessed the mortality status of 1,263 samples and incorrectly assessed 7 (seven) samples, corresponding to an accuracy of 99.45%.

[0029] Similarly, as shown in data table 601, the method of this disclosure (e.g.) Figure 4 The method shown was used to analyze the test data. More specifically, mortality assessment was performed on 250 test samples. The assessment correctly assessed the mortality status of 242 samples and incorrectly assessed 8 samples, corresponding to an accuracy of 96.8%.

[0030] The data associated with the test samples are also graphically illustrated in bar chart 602. More specifically, bar chart 602 is a histogram showing the total number of test samples assigned a given survival probability. For example, approximately 120 samples were assigned a survival probability of 95% to 100%, and approximately 120 samples were assigned a mortality probability of 95% to 100%. However, a relatively small number were assigned a moderate mortality or survival probability. Moderate probabilities are associated with a relatively low level of certainty or confidence in the accuracy of mortality assessment. The data in Table 601 and bar chart 602 demonstrate that this method is capable of producing, for example, logistic regression models with high predictive power for insect mortality rates.

[0031] Figure 7 It shows the relationship with Figure 6The confusion matrix associated with the data is shown in Table 701. Table 701 shows the data derived from the training data. Analysis of the training data correctly predicted 634 insect samples as alive and 639 insect samples as dead. Six live insect samples were incorrectly predicted as dead, and one dead insect sample was incorrectly predicted as alive. Table 702 shows the data derived from the test data. Analysis of the test data correctly predicted 125 insect samples as alive and 125 insect samples as dead. Four live insect samples were incorrectly predicted as dead, and four dead insect samples were incorrectly predicted as alive.

[0032] Figure 8 Examples of what can be combined with Figure 1 The illustrated system is used in an example configuration 800 of a computing device 810. The computing device 810 includes a processor 802 for executing instructions. For example, instructions may be stored in a memory region 804. The processor 802 may include one or more processing units (e.g., in a multi-core configuration).

[0033] Processor 802 is operatively coupled to input / output device 801, enabling computing device 810 to communicate with users (such as laboratory workers or scientific researchers). For example, input / output device 801 can receive and send instructions from the user to perform the methods described above. In some embodiments, computing device 810 can communicate with a user's remote device (such as a mobile tablet computer or smartphone) and can receive and send instructions via a network.

[0034] Input / output device 801 is operatively coupled to digital camera 803. Digital camera 803 is configured to capture digital images of insect specimens for mortality assessment as described herein. For example, digital camera 803 may be combined with, for example, Figure 1 In the system 100 shown, the digital camera 803 is also configured to transmit data associated with captured digital images to an input / output device 801. A processor 802 can receive such data from the input / output device 801 and perform analysis methods on the data, such as... Figure 4 The method is illustrated. In some embodiments, the digital camera 803 communicates directly with the processor 802 and may include internal storage.

[0035] Processor 802 may also be operatively coupled to storage device 806. Storage device 806 is any computer operating hardware suitable for storing and / or retrieving data. In some embodiments, storage device 806 is integrated into computing device 810. For example, computing device 810 may include one or more hard disk drives as storage device 806. In other embodiments, storage device 806 is located external to computing device 810 and is accessible to multiple server computer devices. For example, storage device 806 may include multiple storage units, such as hard disks or solid-state drives, configured in a redundant array of inexpensive disks (RAID). Storage device 806 may include a storage area network (SAN) and / or network attached storage (NAS) system.

[0036] In some implementations, processor 802 is operatively coupled to storage device 806 via storage interface 805. Storage interface 805 is any component capable of providing processor 802 with access to storage device 806. Storage interface 805 may include, for example, an Advanced Technology Attachment (ATA) adapter, a Serial ATA (SATA) adapter, a Small Computer System Interface (SCSI) adapter, a RAID controller, a SAN adapter, a network adapter, and / or any component that provides processor 802 with access to storage device 806.

[0037] Memory region 804 may include, but is not limited to, random access memory (RAM) such as dynamic RAM (DRAM) or static RAM (SRAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), and non-volatile RAM (NVRAM). The above memory types are merely exemplary and therefore do not limit the types of memory that can be used to store computer programs.

[0038] As used herein, the term processor refers to a central processing unit, microprocessor, microcontroller, reduced instruction set circuit (RISC), application-specific integrated circuit (ASIC), logic circuit, and any other circuit or processor capable of performing the functions described herein.

[0039] Figure 9 This is a flowchart of an example method 900 for detecting insect mortality. In an example implementation, part of method 900 may be provided by a system for detecting insect mortality, such as system 100. Figure 1 (As shown) The method 900 is performed. In some embodiments, portions of method 900 may be performed in conjunction with a computing device. In other embodiments, method 900 may include additional, fewer, or alternative actions, including those described elsewhere herein.

[0040] Method 900 begins by placing a specimen (e.g., an insect specimen) 902 under a digital camera. In some embodiments, the insect specimen may be contained within holes in a perforated plate and surrounded by a liquid medium. The system may then be activated in step 904. Activation may include aligning the lens of the digital camera with the holes in the perforated plate. After activation 904, the digital camera captures 906 digital images of the insect specimen. In some embodiments, the digital camera will capture 906 multiple digital images. The digital images may then undergo a preprocessing step 908. In some embodiments, preprocessing step 908 may include removing the outer edges of portions of the digital image that do not include the insect specimen, or removing irrelevant visual data (i.e., image cropping). In other embodiments, preprocessing step 908 may include changing the aspect ratio of the digital image, magnifying, separating portions of the image, or any other process that enables the method to be performed as described herein. The digital images are then analyzed 910 using a Siamese neural network, and the analyzed images are compared to a trained dataset. The viability status of the insect specimen is then determined 912. The results indicating the viability of the insect specimens are then stored in the dataset. In some implementations, the results stored in the dataset can be used to further train the dataset for analyzing other digital images.

[0041] The foregoing written description uses examples to illustrate embodiments of this disclosure and also enables any person skilled in the art to implement this disclosure, including making and using any apparatus or system and performing any combined methods. The patent scope of this disclosure is defined by the claims and may include other examples that would be conceived by a person skilled in the art. Such other examples are also intended to be within the scope of the claims if they have structural elements identical to the literal language of the claims, or if the equivalent structural elements included in such examples are not substantially different from the literal language of the claims.

Claims

1. A system for automatically detecting insect mortality rates, the system comprising: One or more cameras; A removable plate, located at a predetermined distance from the one or more cameras, the removable plate having holes in which insect specimens are disposed; processor; as well as The memory stores instructions that, when executed by the processor, cause the processor to: Transmit one or more signals to the one or more cameras to capture at least two time-series images of the insect specimen; The at least two time-series images of the insect specimens are preprocessed; Using a trained Siamese neural network, the at least two time-series images of the insect specimens are analyzed to determine the viability status of the insect specimens; as well as The results indicating the vitality status of the insect specimen are stored in the memory.

2. The system of claim 1, wherein, in order to preprocess the at least two time-series images of the insect specimen, the memory further stores instructions that, when executed by the processor, cause the processor to crop each of the at least two time-series images of the insect specimen.

3. The system of claim 1, wherein the trained Siamese neural network comprises two convolutional neural networks.

4. The system of claim 3, wherein the two convolutional neural networks are ResNet50 networks.

5. The system of claim 1, wherein, for analyzing the at least two time-series images of the insect specimen, the memory further stores instructions that, when executed by the processor, cause the processor to: The trained Siamese neural network is used to generate at least two sets of embeddings, each set of embeddings uniquely corresponding to one of the at least two time-series images of the insect specimen; Calculate the difference between the at least two sets of embeddings; and The viability state of the insect specimen is determined based on the differences between the at least two sets of embeddings and the classifier layer.

6. The system of claim 5, wherein the classifier layer is a sigmoid classifier layer.

7. The system of claim 1, wherein the trained Siamese neural network is trained on a first set of live specimen images and a second set of dead specimen images, and wherein the first set of images is oversampled compared to the second set of images.

8. The system of claim 7, wherein the first set of live specimen images comprises: One or more original images of the live specimen; as well as An augmented dataset is generated by modifying one or more original images of the live specimen.

9. A method for automatically detecting insect mortality rate, the method comprising: Remove a removable plate located at a predetermined distance from one or more cameras, the removable plate having holes in which insect specimens are placed; Transmit one or more signals to the one or more cameras to capture at least two time-series images of the insect specimen; The at least two time-series images of the insect specimens are preprocessed; The trained Siamese neural network is used to analyze the at least two time-series images of the insect specimens to determine the viability of the insect specimens; as well as The results indicating the vitality status of the insect specimen are stored in a memory coupled to the processor.

10. The method of claim 9, wherein preprocessing the at least two time-series images of the insect specimen includes cropping each of the at least two time-series images of the insect specimen.

11. The method of claim 9, wherein the trained Siamese neural network comprises two convolutional neural networks.

12. The method of claim 11, wherein the two convolutional neural networks are ResNet50 networks.

13. The method of claim 9, wherein analyzing the at least two time-series images of the insect specimen comprises: The trained Siamese neural network is used to generate at least two sets of embeddings, each set of embeddings uniquely corresponding to one of the at least two time-series images of the insect specimen; Calculate the differences between the at least two sets of embeddings; as well as The viability state of the insect specimen is determined based on the differences between the at least two sets of embeddings and the classifier layer.

14. The method of claim 13, wherein the classifier layer is a sigmoid classifier layer.

15. The method of claim 9, wherein the trained Siamese neural network is trained on a first set of live specimen images and a second set of dead specimen images, and wherein the first set of images is oversampled compared to the second set of images.

16. The method of claim 15, wherein the first set of live specimen images comprises: One or more original images of the live specimen; as well as An augmented dataset is generated by modifying one or more original images of the live specimen.

Images