Image Processing Method

Abstract

A method for processing seed images, comprising inputting a seed image comprising at least a portion of a seed into a trained neural network to generate a value associated with the coverage of a coating on the seed, wherein the trained neural network has been trained to generate a value associated with the coverage of a coating on the seed using a plurality of training images, each of which comprises at least a portion of a training seed, and each training image is labeled with a value associated with the coverage of a coating on the training seed.

Description

[Technical Field]

[0001] The present disclosure relates to methods of processing seed images, and particularly, but not exclusively, to methods of processing seed images including coatings. [Background technology]

[0002] Seeds used in agriculture are often coated. Coatings can be provided for a variety of reasons. It is common to provide seeds with coatings to protect them from damage during handling, to prevent dust, and for cosmetic appearance. Such coatings may also provide the advantage of protecting the seeds from pest attack and disease and smoothing the seed surface to facilitate planting. Plant nutrients or other growth stimulants can be incorporated into the seed coating to control seed germination, i.e., germination rate. Plant protection agents, such as pesticides (e.g., fungicides and insecticides), can be incorporated to further protect the seeds from disease and / or pest attack.

[0003] Seeds are often subjected to mechanical stresses, particularly during processing. Accordingly, coatings on the seeds are also subjected to mechanical stresses. It is beneficial for the coating on the seeds to be resistant to abrasion resulting from the mechanical stresses so that the seeds remain sufficiently coated, the coating can perform its intended function, the seeds remain visible in the soil, and / or the coating remains a visible means for identifying the type of seed and its particular coating.

[0004] Abrasion resistance is the ability of a coating to withstand mechanical stress and is an index for evaluating the quality of film coatings on seeds. Visual inspection of seeds can be used to determine the abrasion resistance of seed coatings, particularly by evaluating how coated the seeds are after being exposed to abrasion. However, such methods can be subjective due to many random factors that cannot be adequately controlled (e.g., personal preference, light, or visual reference), resulting in high variability even within the same sample. It is desirable to develop a more accurate and consistent method for evaluating the abrasion resistance of coatings. Summary of the Invention

[0005] According to an embodiment of one aspect of the present invention, there is provided a method for processing seed images, the method comprising inputting a seed image comprising at least a portion of the seed into a trained neural network to generate a value associated with the coverage of a coating on the seed, the trained neural network being trained to generate the value associated with the coverage of the coating on the seed using a plurality of training images, each comprising at least a portion of a training seed, and each training image being labeled with a value associated with the coverage of the coating on the training seed.

[0006] A value related to the coverage of a coating on a seed may indicate the resistance of the coating to abrasion. A value related to the coverage of a coating on a seed / training seed may indicate the percentage of the seed surface that is covered with the coating. A value related to the coverage of a coating on a seed may indicate the uniformity of the coverage of a coating on a seed. A value related to the coverage of a coating on a seed may constitute an indicator of how uniformly the coating is applied to the seed surface and / or how much of the seed surface is covered with the coating. The amount of surface coverage with the coating may indicate the abrasion resistance of the coating. For example, when a coated seed is subjected to mechanical stress, the amount of coating remaining may indicate the abrasion resistance of the coating, or the thickness of the coating may indicate the abrasion resistance of the coating.

[0007] The method may further include training the neural network. The trained neural network may be trained, or may have been trained, using first training images of training seeds including a coating of a first color and second training images of training seeds including a coating of a second color. The inputting step may include inputting seed images of seeds including a coating of the first color.

[0008] A seed image may be generated by extracting a region of a single seed from an image containing multiple seeds, which may be extracted using an algorithm that detects circles in the image to find the area within the single seed, and then the area within the single seed may be extracted.

[0009] The training image may be generated by extracting a region of a single training seed from an image containing multiple training seeds, which may be extracted using an algorithm that detects circles in the image to find the area within the single training seed, and then the area within the single training seed may be extracted.

[0010] The multiple seed images may be generated from the same image or may be taken from the same sample (e.g., the same seed or the same sample of seeds). The sample may be a sample of seeds that have been coated with substantially the same amount (and type) of coating and subjected to substantially the same amount of mechanical stress. These images may be input into a trained neural network to generate multiple values ​​related to the coating coverage corresponding to the seed image. The multiple values ​​may be averaged.

[0011] The neural network may be at least one of a convolutional neural network and a deep neural network. The neural network may include at least one of an input layer, a 2D convolutional layer, a batch normalization layer, a ReLU layer, a max pooling layer, a fully connected layer, and a recurrent layer (each of which).

[0012] According to an embodiment of one aspect of the present invention, there is provided a computer program that, when executed by a computing system including processor hardware and memory hardware, causes the processor hardware to perform any of the methods described herein.

[0013] According to an embodiment of one aspect of the present invention, there is provided an apparatus comprising processor hardware and memory hardware, the memory hardware storing processing instructions that, when executed by the processor hardware, cause the processor hardware to perform any of the methods described herein.

[0014] To avoid unnecessary duplication of effort and repetition in this specification, certain features will be described in relation to only one or a few aspects or embodiments of the invention, however, it will be understood that, where technically possible, a feature described in relation to any aspect or embodiment of the invention can also be used with any other aspect or embodiment of the invention. [Brief explanation of the drawings]

[0015] For a better understanding of the present invention and to show more clearly how the same may be carried into effect, reference will now be made, by way of example, to the accompanying drawings in which:

[0016] [Figure 1a] 1 is a flowchart of a method according to an example. [Figure 1b] 1 is a flowchart of a method according to an example. [Figure 2] 1 is an illustration of an image involved in an image processing method according to an example; [Figure 3] According to one example, an illustration of seed segments for training a neural network is shown, along with their ascribed coverage values ​​(labels). [Figure 4] FIG. 1 is an illustration of an image used for neural network validation. [Figure 5] 1 shows an illustration of a seed portion for neural network validation, along with its ascribed coverage values ​​and the predictions produced by the neural network. [Figure 6] 1 is a graph showing the correlation between imputed coverage values ​​and predictions produced by a neural network. [Figure 7] This is a color version of Figure 2. [Figure 8] This is a color version of Figure 3. [Figure 9] This is a color version of Figure 4. [Figure 10] This is a color version of Figure 5. DETAILED DESCRIPTION OF THE INVENTION

[0017] Seed coatings are typically colored to indicate that the seeds have been treated. The particular color of the coating may indicate the composition of the coating. Alternatively, the color may be selected to make the seeds more visible in the soil. Blue coatings are less noticeable in the soil, while red coatings are more noticeable. Seeds may also be coated with colors that are less attractive to animals. Field crops such as corn and sunflowers are commonly coated with red or blue colors.

[0018] Colored coatings on seeds are useful for determining the abrasion resistance of the coating (or the resistance of the coating to being removed or displaced from the seed when subjected to mechanical stress) because the color contrast between the colored coating and the untreated seed can be used to determine the coverage and / or thickness of the coating on the seed. Thus, seeds can be visually inspected and assigned a value indicative of the coverage of the coating on the seed, thereby indicating the resistance of the coating to abrasion and / or the percentage of the seed surface that is covered with the coating. The value associated with seed coverage may indicate how evenly the coating is applied to the seed's surface and / or how much of the seed's surface is covered with the coating. The amount of surface coverage and / or thickness of the coating may indicate the abrasion resistance of the coating.

[0019] The abrasion resistance of a coating on a seed may be tested by subjecting the coated seed to mechanical stress. Abrasion of the coating on the seed may be achieved, for example, by placing at least one seed on a wheel (drum, carousel) and rotating the wheel for a certain time and / or number of revolutions. The wheel may rotate in a vertical plane about a horizontal axis. Alternatively, the wheel may rotate in a horizontal plane about a vertical axis. Any suitable machine for applying mechanical stress to the seeds may be used. For example, an abrasion testing machine may be used. Examples of such machines include PharmaTest's Models PTF 200, PTF 20E, or PTF 20ER (typically used to test the friability or abrasion of pharmaceutical tablets, but can be used on seeds as well). Abrasion can then be visually assessed by assessing how much of the color coating remains, and a value within a range can be assigned to the coating. The coating may be completely removed from the seed surface, or the thickness of the coating may be reduced (resulting in a coating color or shade that differs from that of the coating at full thickness). Abrasion of seeds (or exposure to mechanical stress) not only causes material loss, such as a decrease in layer thickness or removal of portions of the coating, but can also cause the coating to be displaced to other parts of the seed. As a result, some areas of the coating may be thinned or completely removed, while other areas may have an increased coating layer thickness. This can result in a more uneven coating, which cannot be measured by methods such as gravitational measurement, but can be measured by assessing the coating thickness. Therefore, the abrasion resistance of a coating may be determined by assessing the coating thickness on seeds exposed to mechanical stress. A lower value in the range may indicate no abrasion (high coating coverage), while an upper value in the range may indicate that most of the coating has been removed. For example, a number between 0 (no abrasion / generally complete coating coverage) and 5 (almost no coating remaining) may be assigned to the seed or coating.

[0020] However, visual inspection of seeds can be relatively subjective, with many random factors that cannot be adequately controlled (e.g., personal preference, light or visual reference), potentially resulting in high variability in the determined abrasion resistance even within the same sample.

[0021] As an example, a method for processing seed images is provided, the method comprising inputting a seed image comprising at least a portion of the seed into a trained neural network to generate a value associated with (or relating to) the coverage of a coating on the seed. The trained neural network is a neural network trained to generate a value associated with the coverage of a coating on the seed using a plurality of training images, each comprising at least a portion of a training seed, and each training image being labeled with a value associated with the coverage of the coating on the training seed. Such a method is shown schematically in FIG. 1a, which outlines the steps of inputting a seed image comprising at least a portion of the seed into a trained neural network to generate a value associated with the coverage of a coating on the seed.

[0022] The method may include training a neural network. Such a method is outlined in FIG. 1b. Thus, the method may include training a neural network using a plurality of training images, each including at least a portion of a training seed, to generate a trained neural network to generate a value associated with the coverage of the coating on the seed, where each training image is labeled with a value associated with the coverage of the coating on the training seed (S100). Seed images including at least a portion of the seed may then be input to the trained neural network to generate a value associated with the coverage of the coating on the seed (S102).

[0023] The determined coverage may indicate the percentage of the seed surface that is covered with the coating (compared to the percentage of the seed surface that is free or substantially free of the coating). The determined coverage may indicate how evenly the coating was applied on the seed surface during the coating process. The determined coverage may indicate the uniformity (in thickness) of the coating (e.g., after exposure to mechanical stress). The determined coverage may indicate the abrasion resistance of the coating. As described above, abrasion resistance is the ability of a coating to withstand mechanical stress. The abrasion resistance of a coating on a seed may be tested by subjecting the seed to mechanical stress and determining the amount of coating removal (the amount of coating removed may be the amount of coating that is completely or substantially completely removed, and / or may include the amount of reduction (or increase) in thickness of the coating). Thus, the determined coverage (e.g., percent coverage, reduction (or increase) in thickness, etc.) of the coating on the seed after the seed has been subjected to mechanical stress may indicate the resistance of the coating to abrasion (abrasion resistance). For example, a value related to the coverage of a coating on a seed may be related to the abrasion resistance of the seed.

[0024] Images of the worn seeds (surfaces) may be taken for use in the method. For example, sample seeds may be placed in a wheel and the wheel may be rotated for a certain time or number of revolutions, thereby exposing the seeds to mechanical stress. Images of the seeds may then be taken using an image extraction means such as a camera or scanner.

[0025] For images generated for training (training images), the same type of seeds may be subjected to the same type of coating and rotated with the wheel for different times or numbers of rotations, resulting in a variety of seed coatings. Images of seeds treated in this way may be captured to obtain images of seeds with different coating coverages. Images of the same type of seeds may also be captured of coated seeds that have not been subjected to mechanical stress and raw (uncoated) seeds. The seed images used for training may show seeds with different levels of coating, including coated seeds that have not been subjected to mechanical stress, coated seeds that have been subjected to mechanical stress, and / or raw uncoated seeds. Images may also vary in quality, resolution, and / or light intensity. These images may be used as training images (e.g., training data) for training a neural network. To validate the method, a large number of training images (validation images) may be set aside to, for example, check that the trained neural network assigns the correct coating / wear resistance values ​​to the images.

[0026] Additionally, such images may be taken of seeds having a coating whose abrasion resistance is to be determined. These images (seed images) may be input into a trained neural network. To determine the abrasion resistance of a coating, it may be desirable to establish a set time or number of revolutions for which a sample of coated seeds is rotated on a wheel, so that different samples of seeds (of the same type of seed) having different coatings are subjected to substantially the same mechanical stress, thus making it possible to compare results for different types of coatings (of the same type of seed).

[0027] The captured images of the seeds may be initial images containing multiple seeds. These initial images may be used as input to a neural network or a trained neural network. Alternatively, data regarding individual seeds contained in the initial images may be extracted. For example, an image of a single seed or a portion of a single seed may be extracted from the initial image. This is advantageous because the initial image may contain many individual seeds, and thus extracting individual seeds increases the amount of training data that can be input to the neural network. Furthermore, to determine coating coverage, images of individual seed portions may be grouped based on the sample from which they were generated (e.g., seeds with the same type of coating, from the same batch of coated seeds), and values ​​indicating coverage may be averaged. For example, images of individual seed portions taken from the same initial image or the same sample may be processed by the neural network to each provide a wear resistance value, which may be averaged to generate an average value related to the coating coverage per sample. This may provide a more accurate value of the wear resistance of a particular coating.

[0028] Computer-implemented image processing methods may be used to extract data about seeds in an image. For example, because seeds (such as corn or tomatoes) tend to have roughly circular or elliptical shapes, a circular Hough transform (CHT)-based algorithm for finding circles in an image may be applied to identify the radius and center coordinates of the seeds in the initial image. The perimeter of the circle may be tangent to the seed's edge. Thus, the circle may be within the seed's perimeter. Alternatively, the circle may capture the seed's overlap region, where the circle overlaps at least a portion of the seed. Note that this method can be applied to any type of seed to identify regions within the seed. These identified regions may then be extracted. This approach is beneficial due to its robustness in the presence of noise, occlusion, and varying illumination of the seed. This extraction method may also be applied to seed images that are input into a trained neural network to determine the coverage of a coating on the seed. The results of this method are shown in Figure 2.

[0029] FIG. 2 shows an image 210 of corn, which has been applied (with a red coating) as described above and subjected to mechanical stress. In the example described herein, the seed used is corn, but it is understood that the method applies to images of any type of seed and / or coating color. The above-described method for extracting individual seed regions is applied to the corn image 210 to generate a processed image 212, in which the regions detected as belonging to individual seeds are indicated by circles. Each individual seed part (or collection of individual seed parts) is then extracted. FIG. 2 shows an example of an extracted individual seed region 214. In particular, the circular region shown in processed image 212 has been extracted, and the circular region is an image of an individual seed part. Any number of extracted regions may be used during training of a neural network or as input to a trained neural network.

[0030] Processing the initial images in this manner may generate multiple images of individual seed parts for use as training images. By way of example, processing 500 initial images in this manner may generate approximately 30,000 images of individual seeds to use for training and validation (10% of the images in such a set may be reserved for validation of the method).

[0031] Several such images are shown in Figure 3. Figure 3 shows 25 individual seed portion regions extracted from an initial image. These images are of seed regions with varying color coatings, coverings (thickness and percentage), lighting, etc. In particular, these images are of yellow corn with blue or red coatings. As shown in this figure, each image is associated with a covering value assessed by visual inspection of the seed (the image number is displayed along with a value labeled "Measured," and the measured value is the covering value assessed by visual inspection of the seed). The value may serve as a general value of seed covering in an initial image with multiple seeds, or it may be applied to each extracted seed portion region extracted from the initial image (the same value attributed to the entire image may be applied to each seed portion). The value may also be within a range of values. As an example, a value between 0 (fully coated) and 5 (not coated at all) may be assigned to each training image to indicate the covering of the coating on the seed. While these images are not visible because they are black-and-white reproductions, because these images (used in the present method) are color images, areas where the coating has been removed may appear as the color of the underlying seed. In this particular example, the underlying seed (corn) is yellow and the coating is blue or red. Thus, the proportion of yellow to red or blue in the image may be associated with the coverage of the coating on the seed. Thus, the proportion of the image consisting of a color corresponding to the uncoated seed color and the proportion of the image consisting of a color corresponding to the coating applied to the seed may indicate the coverage of the coating on the seed (e.g., this may be a method of visually assessing the seeds, and such analysis may be used to determine a value associated with the training images). Note that partial removal of the coating, e.g., a decrease (or increase) in coating thickness, may result in a color that is intermediate between the color of the coating and the color of the underlying seed, or the color of the coating may be a different shade; this intermediate color or different shade associated with partial attrition of the coating may be taken into account when generating a value associated with the coverage of the coating.Thus, the value associated with the coating on a seed may represent how much of the seed coating has been removed (or how much remains), ranging from no coating to complete coverage. Complete coverage is when the seed is substantially completely covered with a coating of a particular thickness (within the margin of error). As an example, a value of 2.5 may represent 50% determined seed coverage on a scale of 0 to 5. The value associated with coating coverage may also be an abrasion resistance score. For example, an abrasion resistance score of 2.5 may represent a seed that is approximately 50% covered by the coating. It should be understood that grayscale images may also be used in the methods described herein, in which case the relative darkness of areas within the image may indicate seed coverage.

[0032] In one example, seeds with different color coatings were used. In particular, blue- and red-coated seeds were used, but it is understood that other color coatings or only one color coating may be used to train the neural network. Alternatively, multiple different color seed coatings may be used to train the neural network. The use of different color coated seeds had the surprising effect that the neural network was able to more accurately identify the coating of the first color seeds when trained with both the first and second color seeds than when trained with only the first color seeds.

[0033] When discussing different colors, each color may be defined by its corresponding wavelength interval in the visible light spectrum. For example, the color "red" may be defined as having a wavelength interval of approximately 700 to 635 nm, "orange" as having a wavelength interval of approximately 635 to 590 nm, "yellow" as having a wavelength interval of approximately 590 to 560 nm, "green" as having a wavelength interval of approximately 560 to 520 nm, "cyan" as having a wavelength interval of approximately 520 to 490 nm, "blue" as having a wavelength interval of approximately 490 to 450 nm, and "purple" as having a wavelength interval of approximately 450 to 400 nm. When referring to "different colors," the colors may be colors that fall into different wavelength intervals. Alternatively, "different colors" or "different shades of a color" may refer to colors at the same wavelength interval but with different wavelengths.

[0034] Images of seed regions / portions may be used as input to a trained neural network or for training the neural network. It is understood that an initial image with multiple seeds may alternatively be used as input to a trained neural network or for training the neural network, rather than individual seed images. However, by extracting individual seeds for input, the number of training data for the neural network increases, thereby increasing the likelihood of training the neural network with higher accuracy.

[0035] To train the neural network, the training images described above (which in this example consist of individual seed images) are input into the neural network. The coverage values ​​attributed to each image (e.g., determined by visual inspection) may also be used by the neural network, e.g., as the image's label. The neural network may process the images to estimate the coverage values ​​associated with the images. For example, given a training dataset such as training images, a forward pass may sequentially compute an output at each layer and propagate a function signal forward through the network. At the final output layer, an objective loss function may measure the error between the inferred output and the given label (e.g., the coverage value attributed to the training image). To minimize the training error, a backward pass may backpropagate the error signal using the chain rule and calculate gradients for all weights throughout the neural network. The weight parameters may then be updated using an optimization algorithm based on stochastic gradient descent (SGD).

[0036] The neural network may be a convolutional neural network (CNN). The neural network may be a deep neural network (DNN). The neural network may have multiple layers (an array of layers). For example, the neural network may include an input layer, a 2D convolutional layer, a batch normalization layer, a ReLU layer, a max pooling layer, a fully connected layer, and an output layer, e.g., a recurrent layer. The neural network may include any number of these layers in any suitable order between the input and output layers. The input to the neural network may be the image described above, and the output may be a value related to the coverage of the coating on the seed, such as the abrasion resistance of the coating, the coverage rate or thickness uniformity of the coating, etc. The neural network may include a set of options for training using stochastic gradient descent with momentum. After training, the model may be re-optimized with a new set of options. For example, the initial learning rate may be set to change less in response to the estimated error each time the model weights are updated in the optimization step.

[0037] An exemplary neural network for use in the present method is outlined below.

[0038] The neural network may include an input layer for inputting a 2D image into the network and applying data normalization. The image input layer may be created for color images (e.g., RGB images) based on the image resolution. For example, for a color image with a resolution of 100x100, each with three RGB values, the input layer may include 100x100x3 (30,000) neurons.

[0039] The neural network may also include a first 2D convolutional layer that applies a sliding convolutional filter to the input, convolving the input by moving the filter vertically and horizontally along the input, computing dot products of the weights and the input, and adding a bias term.

[0040] The neural network may also include a first batch normalization layer that normalizes each input channel across a mini-batch. The use of interlayer batch normalization layers may speed up training of the convolutional neural network and reduce sensitivity to network initialization. The batch normalization layer first normalizes the activation of each channel by subtracting the mean of the mini-batch and dividing by the standard deviation of the mini-batch. Next, the layer shifts the input by a learnable offset β and scales the input by a learnable scale factor γ.

[0041] The neural network may also include a first ReLU layer that performs a threshold operation on each element of the input, where all values ​​less than zero are set to zero.

[0042] The neural network may also include a max pooling layer, which performs downsampling by dividing the input into rectangular pooling regions and calculating the maximum value in each region. As an example, the max pooling layer may have a pool size ∇ ...

[0043] The neural network may further comprise a second 2D convolutional layer that is similar in construction to the first layer, except that there may be twice as many layers.

[0044] The neural network may also include a second batch normalization layer, a second ReLU layer, a second max pooling layer, a third batch normalization layer, and a third ReLU layer.

[0045] The neural network may also include a first fully connected layer that multiplies the input by a weight matrix and adds a bias vector. An output size may be specified for the fully connected layer. As an example, the output size may be the same size as the second or third convolutional 2D layer.

[0046] The neural network may include a fourth ReLU layer and a second fully-connected layer. The size of the second fully-connected layer may be 1 (e.g., the value output from the second fully-connected layer may be 1). The output of the second fully-connected layer may be a value related to the coverage (e.g., wear resistance) of the coating of the seed shown in the input image.

[0047] The neural network may also include a recurrent layer that computes a half mean squared error loss for the regression problem.

[0048] A set of options for training a neural network using momentum stochastic gradient descent may be determined. For example, the number of epochs (an epoch corresponding to a full pass of data) may be set to 250. The execution environment, i.e., the hardware resources for training the network, may use a local or remote parallel pool. For example, calculations / modeling and software development may be performed using parallel computing on a computer with multiple GPUs and CPUs, such as 4 GPUs and 12 CPUs. Additionally, the training data may be divided into small batches (mini-batches) of, for example, 100 in size at each epoch. The initial learning rate (how much the model is modified depending on the estimation error each time the model weights are updated) may be set to, for example, 0.001. Regularization may be used to improve model performance. The regularization coefficient may be set to 0.004. Data may be shuffled before each training epoch and before each network validation. Validation data may be used during training to validate the neural network. Validation accuracy and validation loss for the validation data may be calculated during training. The validation data may be images with corresponding values ​​of the coverage set separately from the training data (validation images set separately from the training images). The number of times the validation set loss can be greater than or equal to the previous smallest loss before network training stops may be set to seven. Plots of training progress may be created during training.

[0049] Validation data (data not used to train the neural network, e.g., images) may be used to evaluate how well the neural network assesses seed cover. To do this, the validation data is input into a trained neural network, which outputs a seed cover or abrasion resistance value. The evaluation may then be performed by comparing the abrasion resistance / covering values ​​determined by the trained neural network with the visually determined abrasion resistance / covering values. Validating the neural network may include calculating the root mean square error (RMSE) between the visually determined cover values ​​and the cover values ​​predicted by the neural network. The trained neural network with the lowest RMSE is the neural network with the best options and architecture. Various configurations of neural networks may be implemented, and the trained neural network with the lowest RMSE may be selected as the trained neural network for use in the method.

[0050] Once the neural network has finished training, the model may be re-optimized. Bayesian optimization may be performed to select and adjust the neural network architecture and options by minimizing the classification error on a validation set. A different set of options may be implemented. For example, the set of options may generally use the same values ​​as described in the example above, or different values. For example, the initial learning rate may be reduced by a factor of 10 to 0.0001. By reducing the initial learning rate during re-optimization, the model may change less with each update of the model weights, depending on the estimated error, potentially refining the resulting seed cover value. Validation of the model using validation data (either the same as or different from that described above) may be performed to evaluate the model's performance, e.g., to determine the model's accuracy.

[0051] The neural network outlined above comprises an architecture and options that the inventors have determined to be particularly efficient in calculating the coverage of a coating on a seed.

[0052] As an example, over 500 images (varying in quality, resolution, and light intensity) of coated, abraded, and raw corn seeds were collected for use as training images (Table 1) and used to train a neural network. Additionally, approximately 10% of the images (46) were used as validation images (27 red-coated and 16 blue-coated). Each of these images was visually inspected and assigned a value indicating the amount of coating on the seed and the abrasion resistance of the coating. [Table 1]

[0053] Approximately 50 images were selected to validate the trained neural network. Figure 4 shows exemplary validation images of the corn (yellow) to be evaluated. Figure 4a) shows a seed with a red coating. Because the seed in this image is exposed to mechanical stress, it consists of various shades of red and various shades of yellow. Figure 4b) shows a seed with a red coating, but not exposed to mechanical stress, so it generally consists of the same shade of red coating. Figure 4c) shows an uncoated seed (raw seed), so it is yellow. Figure 4d) shows a seed with a blue coating. Because the seed in this image is exposed to mechanical stress, it consists of various shades of blue and various shades of yellow. Each seed is assigned a value by the neural network related to the coverage of the coating on the seed, and the average of these values ​​is shown as the final score for abrasion resistance. Table 2 below summarizes the abrasion resistance determined by visual inspection of the seed and the abrasion resistance determined by the trained neural network for each of images 4(a)–4(d). The two values ​​correlate within an acceptable range. [Table 2]

[0054] Figure 5 shows 25 example seed parts extracted from images, along with their visually determined wear resistance (labeled "Expected") and wear resistance predicted by the trained neural network (labeled "Predicted"). Figure 5 shows seed regions with various color coatings, coatings (thickness and percentage), and illumination. In particular, these images show yellow corn with blue or red coatings. Figure 6 shows a plot of measured wear resistance versus predicted wear resistance for the validation images, including the seeds shown in Figure 5. In this example, the correlation between measured and predicted wear resistance is 0.9455. More specifically, the correlation for the red coating is 0.91, and the correlation for the blue coating is 0.98. Therefore, it is clear that the trained neural network is sufficiently accurate in assessing wear resistance.

[0055] 7 to 10 are color versions of FIGS. 2 to 5, respectively.

[0056] In any of the above aspects, the various features may be implemented in hardware or as software modules running on one or more processors. Features of one aspect may also be applied to any of the other aspects.

[0057] The present invention also provides a computer program or computer program product for performing any of the methods described herein, and a computer readable medium having stored thereon a program for performing any of the methods described herein. A computer program embodying the present invention may be stored on a computer readable medium or may be in the form of a signal, for example a downloadable data signal provided from an internet website, or in any other form.

[0058] A computing device, such as a data storage server, may embody the present invention or may be used to implement the method of an embodiment of the present invention. The computing device may include a processor and memory. The computing device may also include a network interface for communicating with other computing devices, such as other computing devices of an embodiment of the present invention.

[0059] For example, an embodiment may comprise a network of such computing devices. The computing devices may also include one or more input mechanisms, such as a keyboard or a mouse, and one or more display units, such as a monitor. The components may be connectable to each other via a bus.

[0060] Memory may include computer-readable media, which may refer to a single or multiple media (e.g., centralized or distributed databases and / or associated caches and servers) configured to carry computer-executable instructions and have data structures stored thereon. Computer-executable instructions may include, for example, instructions and data that are accessible by a general-purpose computer, a special-purpose computer, or a special-purpose processing device (e.g., one or more processors) to cause the computer to perform one or more functions or operations. Accordingly, the term "computer-readable storage medium" may include any medium that can store, encode, or carry a set of instructions for execution by a machine, causing the machine to perform any one or more of the methods of the present disclosure. The term "computer-readable storage medium" may accordingly be considered to include, but is not limited to, solid-state memory, optical media, and magnetic media. By way of example, and not limitation, such computer-readable media may include non-transitory computer-readable storage media such as random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disk storage, magnetic disk storage or other magnetic storage devices, flash memory devices (e.g., solid-state memory devices), and the like.

[0061] A processor may be configured to control a computing device and perform processing operations, such as executing code stored in memory to implement the various methods described herein and claimed. The memory may store data that is read from and written to by the processor. As referred to herein, a processor may include one or more general-purpose processing devices, such as a microprocessor, a central processing unit, etc. A processor may include a complex instruction set computing (CISC) microprocessor, a reduced instruction set computing (RISC) microprocessor, a very long instruction word (VLIW) microprocessor, or a processor implementing other instruction sets or combinations of instruction sets. A processor may also include one or more special-purpose processing devices, such as an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), a digital signal processor (DSP), a network processor, etc. In one or more embodiments, a processor is configured to execute instructions to perform the operations and steps discussed herein.

[0062] The display unit may display representations of data stored by the computing device, and may also display cursors, dialog boxes, and screens that allow interaction between a user and programs and data stored on the computing device. The input mechanism allows a user to input data and instructions into the computing device.

[0063] The network interface (network I / F) may be connected to a network such as the Internet, and may be connectable to other such computing devices via the network. The network I / F 997 may control data input to / output from other devices via the network. Other peripheral devices, such as a microphone, speakers, printer, power supply unit, fan, case, scanner, tracker ball, etc., may also be included in the computing device.

[0064] Embodiments of the present disclosure, its various features and advantageous details, will be more fully described with reference to non-limiting examples described and / or illustrated in the drawings and detailed in the following description. It should be noted that the features illustrated in the drawings are not necessarily drawn to scale, and that features of one embodiment may be employed in other embodiments even if not explicitly described herein, as will be recognized by those skilled in the art. Descriptions of well-known components and processing techniques may be omitted so as not to unnecessarily obscure the embodiments of the present disclosure. The examples used herein are intended merely to facilitate an understanding of how embodiments of the present disclosure may be implemented and to enable those skilled in the art to further practice the same. Accordingly, the examples herein should not be construed as limiting the scope of the embodiments of the present disclosure, which scope is defined solely by the appended claims and applicable law.

[0065] It is understood that embodiments of the present disclosure are not limited to the particular methodologies, protocols, devices, apparatus, materials, applications, etc. described herein, as these may vary. It is also understood that the terminology used herein is used only for the purpose of describing particular embodiments, and is not intended to limit the scope of the embodiments as claimed. It should be noted that as used in this specification and the appended claims, the singular forms "a," "an," and "the" include plural references unless the context clearly dictates otherwise.

[0066] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the disclosed embodiments belong. Preferred methods, devices, and materials are described, although any methods and materials similar or equivalent to those described herein can be used in the practice or testing of the embodiments.

[0067] Although only a few exemplary embodiments have been described in detail above, those skilled in the art will readily appreciate that many modifications are possible in the exemplary embodiments without substantially departing from the novel teachings and advantages of the disclosed embodiments. The above-described embodiments of the present invention may be advantageously used independently of any other of the embodiments or in any feasible combination with one or more other of the embodiments.

[0068] Accordingly, all such modifications are intended to be included within the scope of the embodiments of the present disclosure as defined in the following claims. In the claims, means-plus-function phrases are intended to cover the structures described herein as performing the recited function and structural equivalents as well as equivalent structures.

[0069] Furthermore, any reference signs placed between parentheses in one or more claims shall not be construed as limiting the claim. The use of terms such as "comprises" and "comprises" does not exclude the presence of elements or steps other than those listed in any claim or in the specification as a whole. The singular reference of an element does not exclude the plural reference of such elements, and vice versa. One or more of the embodiments may be implemented by means of hardware comprising several different elements. In a device or apparatus claim enumerating several means, several of these means may be embodied by one and the same item of hardware. The mere fact that certain means are recited in mutually different dependent claims does not indicate that a combination of these means cannot be used to advantage.

Claims

1. 1. A method for processing an image of a seed, comprising: inputting a seed image including at least a portion of a seed into a trained neural network to generate a value related to coverage of the coating on the seed. Including, the trained neural network is trained to generate a value associated with coverage of a coating on a seed using a plurality of training images, each of the training images including at least a portion of a training seed, each training image being labeled with a value associated with coverage of a coating on the training seed; The method, wherein the value associated with the coverage of the coating on the seeds and / or training seeds indicates the resistance of the coating to abrasion.

2. 2. The method of claim 1, wherein the value related to the coverage of the coating on the seeds and / or training seeds indicates at least one of the percentage of the seed surface covered with the coating, the uniformity of the coverage of the coating on the seeds.

3. The method of claim 1 , wherein the method further comprises training the neural network.

4. 10. The method of claim 1, wherein the trained neural network is trained using first training images of training seeds that include a coating of a first color and using second training images of training seeds that include a coating of a second color.

5. The method of claim 4 , wherein the inputting step includes inputting a seed image of a seed including the first color coating.

6. The method of claim 1 , wherein the seed image is generated by extracting a region of a single seed from an image containing multiple seeds.

7. 7. The method of claim 6, wherein the region of the single seed is extracted using an algorithm that detects circles in an image to detect areas within the single seed, and then the area within the single seed is extracted.

8. The method of claim 1 , wherein the training images are generated by extracting a region of a training seed from an image containing multiple training seeds.

9. 9. The method of claim 8, wherein the region of the one training seed is extracted using an algorithm that detects circles in an image to detect areas within the one training seed, and then the area within the one training seed is extracted.

10. 10. The method of claim 1, wherein multiple seed images generated from the same image or taken from the same sample are input to the trained neural network to generate multiple values ​​related to coating coverage corresponding to the multiple seed images, and the multiple values ​​are averaged.

11. The method of claim 1 , wherein the neural network is at least one of a convolutional neural network and a deep neural network.

12. 2. The method of claim 1, wherein the neural network comprises at least one of each of an input layer, a 2D convolutional layer, a batch normalization layer, a ReLU layer, a max pooling layer, a fully connected layer, and a recurrent layer.

13. 10. A computer program product that, when executed by a computing system comprising processor hardware and memory hardware, causes the processor hardware to perform the method of claim 1.

14. 10. An apparatus comprising processor hardware and memory hardware, the memory hardware storing processing instructions that, when executed by the processor hardware, cause the processor hardware to perform the method of claim 1.

Images