Chick sex identification using digital image analysis
Through image analysis and machine learning technology, combined with strategic stimulation of the reflex behavior of chicks, automated chick gender identification is achieved, solving the damage and unreliability problems caused by manual manual operations in traditional methods, and improving the identification efficiency and accuracy.
Patent Information
- Application Number
- CN202380082338.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2022-10-18
- Filing Date
- 2023-10-11
- Publication Date
- 2025-06-27
AI Technical Summary
Traditional feather gender identification methods require manual expansion of chick wings, resulting in injury and unreliability, especially in large-scale applications.
By obtaining and processing images of chicks, using machine learning technology to identify gender-related features, and strategically stimulating the chicks' reflex behavior to automatically spread their wings, thereby achieving automated gender identification.
It improves the accuracy, safety and efficiency of chick gender identification, reduces the risk of human error and chick injury, and is suitable for large-scale gender identification tasks.
Smart Images

Figure CN120226058A_ABST
Abstract
Description
[0001] Cross - Reference to Related Applications This application claims the benefit of priority to U.S. Provisional Application No. 63 / 379,956, filed on October 18, 2022, entitled "Chick Sex Identification Using Digital Image Analysis", the entire content of which is incorporated herein by reference. Technical Field
[0002] The present invention generally relates to chick sex identification, and more particularly to feather sex identification using digital image analysis. Background Art
[0003] Chick sex identification is a method of identifying the sex of chicks or other young fowl, such as to separate female chicks or "pullets" from male chicks or "cockerels". Feather sex identification is a way of chick sex identification based on the growth rate of the feathers on the wings of chicks. Specifically, if the primary flight feathers of a chick are longer than its coverts, the chick is identified as female; if the coverts and primary flight feathers are of the same length or the primary flight feathers are shorter than the coverts, the chick is identified as male. Traditional feather sex identification requires trained staff to manually spread the wings of the chicks, which can cause damage to the wings of the chicks. Summary of the Invention
[0004] Embodiments of the present invention provide systems, methods, and computer - readable media having instructions for identifying the sex of chicks. A method includes moving a chick along a path associated with a stimulus for eliciting an open stance of the chick. The method further includes obtaining one or more images of the chick in the open stance, determining one or more sex - related characteristics of the chick at least in part based on the one or more images, and determining the sex of the chick at least in part based on the one or more sex - related characteristics. One or more parts of the method may include any of a variety of machine - learning techniques, such as supervised machine learning and / or unsupervised machine learning. For example, a supervised machine - learning method can be used to train a neural network to recognize the posture of the chick, identify sex - related characteristics associated with the wings of the chick, and / or evaluate sex - related characteristics for sex determination. Brief Description of the Drawings
[0005] In the drawings, reference numerals may be reused to indicate corresponding between the labeled elements. The drawings are provided to illustrate embodiments of the present invention but do not limit its scope.
[0006] Figure 1A is a schematic diagram showing an example of training a machine - learning model related to the present invention.
[0007] Figure 1B is a schematic diagram showing an example of applying the trained machine - learning model to new observations associated with identifying the sex of chicks.
[0008] Figure 2A Shows a chick gender identification system for accurately identifying the gender of chicks according to some embodiments of the inventive concept.
[0009] Figure 2B - 2F Shows an example comparison of the wing feathers of male and female chicks.
[0010] Figure 3 Is a flowchart of an example program for determining the gender of chicks according to an example embodiment.
[0011] Figure 4A - 4F Shows a plurality of example post - processed images of a chick moving along an example path.
[0012] Figure 5A - 5F Shows a plurality of example post - processed images of a chick moving along an example path.
[0013] Figure 6 Displays a bar chart showing an example relationship between chick age and gender identification accuracy. Detailed Description
[0014] For the purposes of the present invention, the term "chick" is used to generally refer to any breed of chicken. In some cases, the term "chick" typically refers to a young chick less than one day (24 hours) old. In other cases, the term "chick" refers to a chick less than two days (48 hours), one week, or four weeks old. Although the present invention generally relates to identifying the gender of chicks, it should be understood that similar gender identification techniques can be applied to chickens of any age, any bird, or any other animal having gender - related characteristics with identifiable phenotypic features that can indicate gender.
[0015] For the purposes of the present invention, the term "pose" as used herein is a broad term encompassing its ordinary and everyday meaning and can refer to, without limitation, position, orientation, a combination of position and orientation, or any other suitable position information. By way of non - limiting example, the pose of a chick can refer to the position and / or orientation of the chick and / or the position and / or orientation of one or more features of the chick (such as the wings of the chick).
[0016] It is desirable to separate male and female chicks in a hatchery so that they can be managed according to their different needs. Feather sexing is a method of determining the gender of newly hatched chicks based on the growth rate of the wing feathers of the chicks. Using this method, if the primary flight feathers (also known as "main feathers") of a chick are longer than its coverts (also known as "cover feathers"), the chick is identified as female; if the coverts and primary flight feathers are of the same length or the main feathers are shorter than the coverts, the chick is identified as male.
[0017] When chicks rest, they usually fold their wings and keep them close to their body sides (which is generally referred to as "closed stance" in this article). Therefore, traditional feather sexing of chicks usually requires trained staff to manually unfold the wings of the chicks for easy inspection of the coverts and primary flight feathers. Manually unfolding the wings can cause physical damage to the wings of the chicks. In addition, this human-based feather sexing method is usually unreliable (e.g., due to visual fatigue), and there will be problems in large-scale applications where the number of chicks to be sexed is in the hundreds or thousands.
[0018] To address these or other problems, a chick sexing system according to some embodiments of the inventive concept can automatically and accurately identify the sex of a chick by obtaining and processing one or more images of the chick. Specifically, the avian sexing system obtains multiple images of the chick and processes these images to identify images in which the posture of the chick is satisfactory for determining the sex (e.g., the image depicts the wings of the chick being extended). In addition, the avian sexing system can utilize machine learning techniques to identify sex-related features from the wings of the chick and make a sex determination. In addition, in some embodiments, the avian sexing system can automatically classify the chick according to the sex indication of the chick.
[0019] As described above, chicks usually maintain a closed stance, which increases the difficulty of obtaining images that can be used for sexing. To address these or other challenges, the avian sexing system can strategically stimulate the reflex behavior of the chick to cause it to unfold or extend one or both wings. For example, the avian sexing system can include a predetermined path associated with one or more stimuli (such as descent, ascent, speed change, air flow, noise, etc.) designed to trigger a reflex into an open stance. For the purposes of the present invention, the term "descent" is used to generally refer to a change in height or acceleration or both, which causes the chick to at least temporarily deviate from the predetermined path. In addition, for the purposes of the present invention, "open stance" is used to generally refer to any posture of the chick other than the closed stance. The term "open stance" can generally refer to a posture in which one or both wings are fully extended. Additionally or alternatively, the term "open stance" can generally refer to a posture in which one or both wings are partially extended. Non-limitingly, the open stance can refer to a posture in which at least one wing of the chick is extended by at least X% (e.g., 30%, 50%, 75%, 90%) of full extension. The avian sexing system can obtain multiple consecutive images depicting the chick as it moves along the path and responds to the stimuli. In this way, the avian sexing system can increase the likelihood of obtaining one or more images depicting the chick in a satisfactory posture for determining its sex.
[0020] As described herein, it can be understood that the embodiments disclosed herein can significantly improve the throughput, accuracy, and safety associated with chick sex determination. Specifically, the embodiments disclosed herein enable a bird sex determination system to efficiently and accurately determine the sex of hundreds or thousands of chicks, in stark contrast to manual techniques that are typically unreliable in large-scale use due to visual fatigue and other problems associated with human error, or that are not ideal due to issues with employee training and maintenance. Additionally, by strategically stimulating the reflexive behavior of chicks to transition them to an open posture, the bird sex determination system advantageously reduces the likelihood of chick injury. The ability to stimulate the reflexive behavior of chicks to cause their wings to at least partially extend, automatically capture images of the chicks with extended wings, and automatically process the images to determine the sex of the chicks enables a potentially more efficient and safer execution of feather sex determination by: increasing the speed / productivity at which sex determination is performed, reducing the likelihood of chick injury by having the chicks extend their wings on their own rather than manually, increasing the accuracy of feather sex determination by reducing errors associated with human intervention, etc.
[0021] Accordingly, the embodiments disclosed by the present invention are at least improved in terms of video imaging and digital image analysis. Additionally, the embodiments disclosed by the present invention address the technical problems inherent in the industry. These technical problems are solved by the various technical solutions described herein, including causing chicks to extend their wings without human contact, capturing images of the extended wings, selecting images including satisfactory chick postures, analyzing the images to identify sex-related features, determining the sex of the chicks, and / or automatically classifying the chicks based on an indication of their sex. Accordingly, this application generally represents a substantial improvement over existing systems.
[0022] Figure 1A FIG. 7 is a schematic diagram showing an example of training a machine learning model 100 related to the present invention. The machine learning model training described herein can be performed using a machine learning system. The machine learning system can include, or be included in, a computing device, a server, a cloud computing environment, etc., such as the bird sex determination system 200 described in more detail herein.
[0023] As shown by reference numeral 105, a machine learning model can be trained using a set of observations. The set of observations can be obtained from and / or input from historical data (e.g., data collected during one or more of the processes described herein). For example, the set of observations can include data collected from the bird sex determination system 200, as described elsewhere herein. In some embodiments, the machine learning system can receive the set of observations (e.g., as an input) from the bird sex determination system 200 or a storage device.
[0024] As shown by reference numeral 110, a set of observations can be obtained from which a feature set can be observed. The feature set can include a set of variables. The variables can be referred to as features. A particular observation can include a set of variable values corresponding to the set of variables. A set of variable values can be specific to the observation. In some cases, different observations may be associated with different sets of variable values, sometimes referred to as feature values.
[0025] In some embodiments, based on the input received from the avian sex determination system 200, the machine learning system can determine the variables of a set of observations and / or the variable values of a particular observation. For example, the machine learning system can identify a feature set (e.g., one or more features and / or corresponding feature values) from structured data input to the machine learning system, such as by extracting data from specific columns of a table, from specific fields of a form and / or message, and / or by extracting data received in a structured data format. Additionally or alternatively, the machine learning system can receive input from an operator to determine features and / or feature values.
[0026] In some embodiments, the machine learning system can perform natural language processing and / or other feature recognition techniques to extract features (e.g., variables) and / or feature values (e.g., variable values) from text (e.g., unstructured data) input to the machine learning system, such as by identifying keywords and / or values associated with those keywords from the text.
[0027] As an example, a feature set of a set of observations can include a first feature "wing presence", a second feature "visible feathers", a third feature "feather pattern", etc. As shown, for a first observation, the value of the first feature can be "yes", the value of the second feature can be "no", and the value of the third feature can be "primary feathers < coverts", etc. These features and feature values are provided only as examples and may be different in other examples. For example, the feature set can include one or more of the following features: bird posture, wing posture, covert length, primary feather length, relative length of coverts compared to primary feathers, etc. In some embodiments, the machine learning system can preprocess and / or perform dimensionality reduction to reduce the feature set and / or combine the features of the feature set into a minimal feature set. The machine learning model can be trained on the minimal feature set, thereby saving resources (e.g., processing resources and / or memory resources) of the machine learning system used to train the machine learning model.
[0028] This set of observations can be associated with a target variable 115. The target variable 115 can represent a variable with a numerical value (such as an integer value or a floating-point value), a variable with a numerical value that falls within a range of values or has some discrete possible values, a variable that can be selected from one of multiple options (such as one of multiple classes, classifications, or labels), or a variable with a boolean value (such as 0 or 1, true or false, yes or no, male or female), etc. The target variable can be associated with a target variable value, and the target variable value can be specific to the observation. In some cases, different observations may be associated with different target variable values. In Example 100, the target variable 115 is the gender of a bird, and the value it has for the first observation is "male".
[0029] The above feature set and target variable are provided only as examples, and other examples may be different from the above. For example, for the target variable "gender", the feature set can include one or more of the length of the coverts, the length of the primary flight feathers, etc.
[0030] The target variable can represent the value that a machine learning model is trained to predict, and the feature set can represent the variables that are input into the trained machine learning model to predict the value of the target variable. This set of observations can include the target variable values such that a machine learning model can be trained to identify the features in the feature set that give rise to the target variable values. A machine learning model trained to predict the target variable values can be referred to as a supervised learning model or a prediction model. When the target variable is associated with continuous target variable values (such as a series of numbers), the machine learning model can employ regression techniques. When the target variable is associated with categorical target variable values (such as classes or labels), the machine learning model can employ classification techniques.
[0031] In some embodiments, the machine learning model can be trained on a set of observations that do not include the target variable (or include the target variable, but the machine learning model is not being executed to predict the target variable). This can be referred to as an unsupervised learning model, an automated data analysis model, or an automated signal extraction model. In such a case, the machine learning model can learn various features from the set of observations without labels or supervision, and can provide an output indicating such features, for example, by using clustering and / or association to identify relevant groups of items in the set of observations.
[0032] As further shown, the machine learning system can divide the set of observations into a training set 120 and a test set 125. The training set 120 includes a first subset of observations of the set of observations, and the test set 125 includes a second subset of observations of the set of observations. The training set 120 can be used to train (e.g., fit or tune) a machine learning model, and the test set 125 can be used to evaluate the machine learning model trained using the training set 120. For example, for supervised learning, the test set 125 can use the first subset of observations for initial model training, and the test set 125 can be used to test whether the trained model accurately predicts the target variable in the second subset of observations. In some embodiments, the machine learning system can divide the set of observations into the training set 120 and the test set 125 by including a first portion or first percentage of the set of observations in the training set 120 (e.g., 75%, 80%, or 85%, etc.) and including a second portion or second percentage of the set of observations in the test set 125 (e.g., 25%, 20%, or 15%, etc.). In some embodiments, the machine learning system can randomly select the observations to be included in the training set 120 and / or the test set 125.
[0033] As indicated by reference numeral 130, the machine learning system can use the training set 120 to train a machine learning model. Such training can include the machine learning system executing a machine learning algorithm to determine a set of model parameters based on the training set 120. In some embodiments, the machine learning algorithm can include a regression algorithm (e.g., linear regression or logistic regression), and the regression algorithm can include a regularized regression algorithm (e.g., lasso regression, ridge regression, or elastic net regression). Additionally or alternatively, the machine learning algorithm can include a decision tree algorithm, and the decision tree algorithm can include an ensemble of trees algorithm (e.g., generated using bagging and / or boosting), a random forest algorithm, or a boosting tree algorithm. The model parameters can include the attributes of the machine learning model that learn from the data input into the model (e.g., the training set 120). For example, for a regression algorithm, the model parameters can include regression coefficients (e.g., weights). For a decision tree algorithm, by way of example, the model parameters can include the decision tree split locations.
[0034] As shown by reference numeral 135, the machine learning system may use one or more hyperparameter sets 140 to adjust the machine learning model. Hyperparameters may include structural parameters that control the machine learning system's execution of a machine learning algorithm, such as constraints imposed on the machine learning algorithm. Different from model parameters, hyperparameters do not learn from the data input into the model. Example hyperparameters of a regularization regression algorithm include the penalty intensity (e.g., weight) applied to the regression coefficients to mitigate overfitting of the machine learning model to the training set 120. The penalty may be applied based on the magnitude of the coefficient values (e.g., for lasso regression, e.g., penalizing large coefficient values), may be applied based on the squared magnitude of the coefficient values (e.g., for ridge regression, e.g., penalizing large squared coefficient values), may be applied based on the ratio of the magnitude to the squared magnitude (e.g., for elastic net regression), and / or may be applied by setting one or more feature values to zero (e.g., for automatic feature selection). Example hyperparameters of a decision tree algorithm include the tree ensemble techniques to be applied (e.g., bagging, boosting, random forest algorithm, and / or boosted tree algorithm), the number of features to be evaluated, the number of observations to be used, the maximum depth of each decision tree (e.g., the number of branches allowed for a decision tree), or the number of decision trees to be included in a random forest algorithm.
[0035] To train a machine learning model, the machine learning system may identify a set of machine learning algorithms to be trained (e.g., based on operator input identifying one or more machine learning algorithms and / or based on a random selection of a set of machine learning algorithms), and may use the training set 120 to train the set of machine learning algorithms (e.g., independently for each machine learning algorithm in the set). The machine learning system may use one or more hyperparameter sets 140 to adjust each machine learning algorithm (e.g., based on operator input identifying the hyperparameter set 140 to be used and / or based on randomly generating hyperparameter values). The machine learning system may use a specific machine learning algorithm and the corresponding hyperparameter set 140 to train a specific machine learning model. In some embodiments, the machine learning system may train multiple machine learning models to generate a set of model parameters for each machine learning model, where each machine learning model corresponds to a different combination of a machine learning algorithm and the hyperparameter set 140 of that machine learning algorithm.
[0036] In some embodiments, when training a machine learning model, a machine learning system may perform cross-validation. Cross-validation can be used to obtain a reliable assessment of the machine learning model performance using only the training set 120 without using the test set 125, for example, by dividing the training set 120 into multiple groups (e.g., based on operator input identifying the number of groups and / or based on randomly selecting multiple groups), and using these groups to evaluate the model performance. For example, using k-fold cross-validation, the observations in the training set 120 can be divided into k groups (e.g., in order or randomly). For the training process, one group can be labeled as the held-out group and the remaining groups can be labeled as the training groups. For the training process, the machine learning system can train the machine learning model on the training groups and then test the machine learning model on the held-out group to generate a cross-validation score. The machine learning system can repeat this training process with different held-out groups and different test groups to generate cross-validation scores for each training process. In some embodiments, the machine learning system can train the machine learning model independently k times, with each individual group serving as the held-out group once and as a training group k - 1 times. The machine learning system can combine the cross-validation scores for each training process to generate an overall cross-validation score for the machine learning model. The overall cross-validation score can include, for example, an average cross-validation score (e.g., across all training processes), a standard deviation across the cross-validation scores, or a standard error across the cross-validation scores.
[0037] In some embodiments, when training a machine learning model, a machine learning system may perform cross-validation by dividing the training set into multiple groups (e.g., based on operator input identifying the number of groups and / or based on randomly selecting multiple groups). The machine learning system may perform multiple training processes and may generate cross-validation scores for each training process. The machine learning system may generate an overall cross-validation score for each set of hyperparameters 140 associated with a particular machine learning algorithm. The machine learning system may compare the overall cross-validation scores for different sets of hyperparameters 140 associated with a particular machine learning algorithm and may select the set of hyperparameters 140 with the best overall cross-validation score (e.g., highest accuracy, lowest error, or closest to a desired threshold) for training the machine learning model. Then, the machine learning system may train the machine learning model using the selected set of hyperparameters 140 without cross-validation (e.g., using all the data in the training set 120 without any held-out groups) to generate a single machine learning model for the particular machine learning algorithm. Then, the machine learning system may test the machine learning model using the test set 125 to generate a performance score, such as mean squared error (e.g., for regression), mean absolute error (e.g., for regression), or area under the receiver operating characteristic curve (e.g., for classification). As described below in connection with Figure 1BIf the machine learning model performs well (e.g., the performance score meets a threshold), the machine learning system can store the machine learning model as the trained machine learning model 145 for analyzing new observations.
[0038] In some embodiments, the machine learning system can perform cross-validation as described above on multiple machine learning algorithms (e.g., regularized regression algorithms, different types of regularized regression algorithms, decision tree algorithms, or different types of decision tree algorithms) (e.g., independently). Based on performing cross-validation on multiple machine learning algorithms, the machine learning system can generate multiple machine learning models, where each machine learning model has the best overall cross-validation score for the corresponding machine learning algorithm. Then, the machine learning system can train each machine learning model using the entire training set 120 (e.g., without cross-validation) and can test each machine learning model using the test set 125 to generate the corresponding performance score for each machine learning model. The machine learning model can compare the performance scores of each machine learning model and can select the machine learning model with the best performance score (e.g., the highest accuracy, the lowest error, or the closest to the desired threshold) as the trained machine learning model 145.
[0039] As described above, Figure 1A is provided only as an example. Other examples may be different from those Figure 1A described in combination. For example, the machine learning model can be trained using a different process from that Figure 1A described in combination. Additionally or alternatively, the machine learning model can employ different machine learning algorithms from those Figure 1A described in combination, such as Bayesian estimation algorithms, k-nearest neighbor algorithms, prior algorithms, k-means algorithms, support vector machine algorithms, neural network algorithms (e.g., convolutional neural network algorithms), and / or deep learning algorithms.
[0040] Figure 1B is a schematic diagram showing an example of applying the trained machine learning model to a new observation associated with identifying the sex of chicks. The new observation can be input into the machine learning system that stores the trained machine learning model 145 (e.g., the trained machine learning model 145 as described in combination Figure 1A ). The machine learning system can include or be included in a computing device, a server, or a cloud computing environment, such as the avian sex identification system 200 of FIG. 2.
[0041] As shown by reference numeral 160, a machine learning system can receive a new observation (or a set of new observations) and can input the new observation into a machine learning model. As shown, the new observation can include, by way of example, a first feature "Is there a wing?", a second feature "Are feathers visible?", a third feature "Feather pattern", etc. The machine learning system can apply the trained machine learning model 145 to the new observation to generate an output 170 (e.g., a result). The type of output may depend on the type of machine learning model and / or the type of machine learning task being performed. For example, the output 170 can include a predicted (e.g., estimated) value of a target variable (e.g., a value within a continuous value range, a discrete value, a label, a category, or a classification), such as when using supervised learning. Additionally or alternatively, the output 170 can include information identifying the cluster to which the new observation belongs and / or information indicating the similarity between the new observation and one or more previous observations (e.g., which can be new observations previously input into the machine learning model and / or observations used to train the machine learning model), such as when using unsupervised learning.
[0042] In some embodiments, the trained machine learning model 145 can predict that the value of the target variable for the gender of the new observation is "female", as shown by reference numeral 180. Based on this prediction (e.g., based on a value with a specific label or classification, or based on a value that meets or does not meet a threshold), the machine learning system can provide a recommendation and / or an output for determining the recommendation, such as providing an indication that the chick is female. Additionally or alternatively, the machine learning system can perform an automated operation and / or can cause an automated operation to be performed (e.g., by instructing another device to perform the automated operation), such as automatically sorting the chick into a female bin. As another example, if the machine learning system predicts that the value of the target variable for gender is "unknown" or a similar value, the machine learning system can provide a different recommendation (e.g., "Manually check the gender") and / or perform or cause to be performed a different automated operation (e.g., automatically sorting the chick into an "unknown" bin). In some embodiments, the recommendation and / or the automated operation can be based on the value of the target variable having a specific label (e.g., a classification or a category) and / or can be based on whether the value of the target variable meets one or more thresholds (e.g., whether the value of the target variable is greater than a threshold, less than a threshold, equal to a threshold, or falls within a threshold range).
[0043] In this way, the machine learning system can apply a rigorous and automated process to determine the gender of a chick. As opposed to allocating the resources (e.g., computational or manual) required for dozens, hundreds, or thousands of operators to manually determine the gender using features or feature values, the machine learning system is capable of identifying and / or characterizing dozens, hundreds, thousands, or millions of features and / or feature values for dozens, hundreds, thousands, or millions of observations, thereby improving accuracy and consistency and reducing the latency associated with chick gender identification.
[0044] As described above, Figure 1B is provided only as an example. Other examples may be different from those combined Figure 1B described.
[0045] Figure 2A FIG. 200 shows a bird sex identification system 200 for accurately identifying the sex of chicks according to some embodiments of the inventive concept. The bird sex identification system 200 includes a path system 210, an imaging system 220, a posture management system 230, a sex indicator 240, and a classification system 250. For simplicity of discussion and without limiting the present invention, Figure 2A only one path system 210, imaging system 220, posture management system 230, sex indicator 240, and classification system 250 are shown, but multiple ones may be used. In addition, any one of these systems / components may be combined or further separated without departing from the scope of the inventive concept.
[0046] Any of the above devices, components, or systems of the bird sex identification system 200 may communicate via a network (not shown). For example, the network may include any type of communication network, such as one or more of a wide area network (WAN), a local area network (LAN), a cellular network (e.g., LTE, HSPA, 3G, and other cellular technologies), an ad hoc network, a satellite network, a wired network, a wireless network, etc. In some embodiments, the network may include the Internet. In addition, it should be understood that any two or more of the path system 210, imaging system 220, posture management system 230, sex indicator 240, and classification system 250 may be combined with each other or may be separated from the bird sex identification system 200.
[0047] The path system 210 facilitates guiding, positioning, and / or orienting the chicks as part of the sex identification process performed by the bird sex identification system 200. The path system 210 may define a path along which one or more chicks may move. The path may include any physical structure that defines a generally pre-planned path or trajectory. For example, the path may include a static structure (e.g., a chute) having an inclined surface along which the chicks may navigate or slide to move along the path. As another example, the path may include a dynamic structure (e.g., a conveyor belt) that transports the chicks along the path. As another example, the path may include any path or trajectory for guiding or sorting the chicks to a desired location.
[0048] The path may include one or more stimuli that prompt or trigger the chick to transition to an open posture. For example, a combination of one or more of the speed of movement along the path, the slope of the path, and / or the sense of falling generated by moving along the path may cause the chick to at least temporarily extend its wings. In cases where the path includes a chute, the chute may include one or more sections of height descent or acceleration increase / decrease that increase the likelihood of the chick reflexively extending its wings. Similarly, in cases where the path includes a conveyor belt, the route of the conveyor belt may include one or more sections of height descent, belt vibration, or speed increase / decrease that increase the likelihood of the chick reflexively extending its wings. As another example, the path system 210 may include at least one stimulating device along the path for causing the chick to reflexively extend its wings. The stimulating device may include, but is not limited to, a fluid sprayer (e.g., for spraying a puff of air or water) or a noise maker. In some cases, the stimulating device advantageously causes the chick to extend or stretch at least one of its wings, for example, by giving the chick a gentle startle.
[0049] As described, the stimuli may increase the likelihood of the chick stretching at least one of its wings; however, in some cases, the stimuli may not actually cause the chick to stretch at least one of its wings. Thus, to further increase the likelihood that the chick will stretch at least one of its wings at least once during movement along the path, the path system 210 may include multiple stimuli and / or sustained stimuli (such as a long height descent). In some embodiments, at least two of the stimuli are different types of stimuli (such as height descent, ascent, speed increase, speed decrease, fluid spraying, etc.) to increase the likelihood that the chick will transition to an open posture at least once while moving along the path.
[0050] Similarly, one or more stimuli may cause the chick to be positioned and / or oriented in a satisfactory posture. As described, the posture of the chick affects whether an image of the chick can be used for reliable sex determination. Thus, for the path system 210, it is advantageous to promote a satisfactory posture. As described in more detail below, a satisfactory posture may include any posture from which a reliable sex indication (e.g., accuracy rate X%) can be determined. For example, a satisfactory posture may include, but is not limited to: the chick facing a particular direction (e.g., towards the camera, away from the camera, etc.), a threshold portion of the chick depicted in the image (e.g., greater than 50%, 35%, etc.), the chick not being hunched or stooped, or the wings of the chick facing a particular direction (e.g., towards the camera, away from the camera, etc.).
[0051] Imaging system 220 may obtain one or more images (or video streams) of one or more chicks. For example, imaging system 220 may include one or more cameras configured to capture one or more images and / or may include a processor configured to obtain one or more images (e.g., from local or remote storage). In some cases, imaging system 220 obtains a plurality of images corresponding to discrete time periods during which a chick moves along a path defined by path system 210. In some embodiments, imaging system 220 may capture images of a scene including a chick at a predetermined time interval (e.g., X milliseconds, X seconds, etc.). For example, the plurality of images may be time series images corresponding to a frame rate of X frames per second. Additionally or alternatively, the plurality of images may correspond to moments when a chick may be in a satisfactory posture, such as moments associated with a stimulus designed to prompt or trigger a chick to spread or extend at least one of its wings. By obtaining a plurality of images corresponding to discrete time periods during which a chick moves along a path, imaging system 220 advantageously increases the likelihood of obtaining an image from which an accurate sex indication (e.g., accuracy rate of X%) can be determined.
[0052] Imaging system 220 may capture images. For example, imaging system 220 may include one or more image capture devices, which may include one or more types of cameras and / or lighting devices, including but not limited to conventional imaging systems utilizing RGB and / or grayscale cameras with broad-spectrum (e.g., white light) illumination. As another example, one or more image capture devices may include dedicated hardware (e.g., to enhance the contrast of feathers). In some cases, imaging system 220 may be implemented as an Imaging Development System (IDS) camera. For example, IDS cameras may be high-performance, easy-to-operate USB, GigE, and 3D cameras with a variety of sensors and models, and / or IDS cameras may be special types of cameras suitable for working in harsh conditions (e.g., high temperature, high pressure, and vibration). As another example, imaging system 220 may be implemented as an OAK camera. For example, an OAK camera may include OAK API software and one or more different types of hardware (e.g., OAK-1 and OAK-D). An OAK camera may be a miniature artificial intelligence (AI) and computer vision (CV) engine, where OAK-D provides spatial AI utilizing stereo depth in addition to the 4K / 30 12MP camera common to both models.
[0053] Imaging system 220 may store all captured images. Alternatively, imaging system 220 may store only certain selected captured images. For example, imaging system 220 may process or preprocess images in real time or near real time to identify which images (if any) include chicks with a described satisfactory pose. Imaging system 220 may capture images of chicks as they move across the path. Multiple images may be transmitted to pose management system 230 and / or gender indicator 240 for image processing (e.g., using machine learning algorithms) and / or display. The multiple images may include a series of images (e.g., + / - offset) taken within a predetermined time period (e.g., the time period during which the chick moves along the path defined by path system 210).
[0054] Pose management system 230 may obtain images from imaging system 220 and may process the images (e.g., in real time or near real time) to determine whether the images include a chick and / or whether the pose of the chick meets pose criteria. In this way, pose management system 230 may evaluate the images to determine whether the images may be used to accurately determine the gender of the chick. Pose management system 230 may evaluate the images to determine whether the images are satisfactory. In some embodiments, at least in part based on a determination that the images may be used to accurately determine the gender of the chick or a determination that the chick in the images meets the pose criteria, pose management system 230 determines that the images are satisfactory. Necessarily, at least in part based on a determination that the images cannot be used to accurately determine the gender of the chick or a determination that the chick in the images does not meet the pose criteria, pose management system 230 may determine that the images are not satisfactory. In some cases, if the bird is located at a specific location in path system 210 (e.g., about 1 / 3 down the chute and about 2 inches from the end of the chute) or within a specific field of view of the camera, pose management system 230 determines that the images are satisfactory.
[0055] Pose management system 230 may use various criteria, which may be referred to as "pose criteria", to determine whether the images are satisfactory. For example, if the pose of the chick meets the pose criteria, pose management system 230 may determine that the images are satisfactory. Necessarily, if the pose of the chick does not meet the pose criteria, pose management system 230 may determine that the images are not satisfactory. The pose criteria may include, but are not limited to, one or more body pose thresholds, body orientation thresholds, body position thresholds, wing pose thresholds, wing orientation thresholds, wing position thresholds, body presence thresholds, etc. The pose criteria may dynamically determine whether the images are satisfactory based on any one or any combination of the following: body pose, wing pose, visibility of the hidden wing and / or primary wing, etc.
[0056] As an example, the pose criteria may include a body pose threshold. The body pose threshold may correspond to the body pose of a chick that facilitates sex determination. For example, at least partially based on the determination that the chick is fully or substantially captured within the image, the pose management system 230 may determine that the body pose threshold is satisfactory. As another example, at least partially based on the determination that the chick is standing (e.g., rather than bent or hunched), the pose management system 230 may determine that the body pose threshold is satisfactory. As another example, at least partially based on the determination that the chick is facing a specific direction (e.g., facing the camera, facing away from the camera) such that the wings face the camera, etc., the pose management system 230 may determine that the body pose threshold is satisfactory. Inevitably, at least partially based on the determination that the chick is partially captured within the image, the chick is bent or hunched, or the chick is facing a specific direction (e.g., facing away from the camera) such that the wings do not face the camera, etc., the pose management system 230 may determine that the body pose threshold is not satisfactory.
[0057] As another example, the pose criteria may include a body orientation threshold. For example, at least partially based on the determination that the orientation of the chick in the image facilitates accurate sex determination, the pose management system 230 may determine that the body orientation threshold is satisfactory. As another example, the pose criteria may include a body position threshold. For example, at least partially based on the determination that the position of the chick in the image facilitates accurate sex determination, the pose management system 230 may determine that the body position threshold is satisfactory.
[0058] As another example, the pose criteria may include a wing pose threshold, a wing orientation threshold, or a wing position threshold. For example, at least partially based on the determination that the pose (e.g., position and / or orientation) of at least one wing of the chick in the image facilitates accurate sex determination, the pose management system 230 may determine that the wing pose threshold, the wing orientation threshold, or the wing position threshold is satisfactory. For example, at least partially based on the presence of an extended (or partially extended) wing, the wing pose threshold may be determined to be satisfactory. As another example, at least partially based on the presence of coverts and / or primary flight feathers on the wing or the ability to identify the coverts and / or primary flight feathers on the wing, the wing pose threshold may be determined to be satisfactory. Inevitably, at least partially based on the absence of an extended (or partially extended) wing or the absence or inability to identify the coverts and / or primary flight feathers on the wing, the wing pose threshold may be determined to be not satisfactory. The pose management system 230 may include a wing detection module 232 (e.g., a machine learning model / algorithm) that helps to determine the presence of an extended wing of the chick, as wings kept close to the body are not suitable for feather detection or identification.
[0059] If the pose management system 230 determines that the image is unsatisfactory (e.g., at least in part based on pose criteria), the pose management system 230 can make one or more decisions. For example, the pose management system 230 can determine that the image should not be evaluated to determine the sex of the chick. Thus, the pose management system 230 can determine whether to discard, ignore, or not select the image or data associated with the image. As another example, the pose management system 230 can determine that a sex determination associated with the image should be given a lower weight value or confidence parameter than a sex determination associated with a satisfactory image.
[0060] If the pose management system 230 determines that the image is satisfactory (e.g., at least in part based on pose criteria), the pose management system 230 can make one or more decisions. For example, the pose management system 230 can determine that the image should be evaluated to determine the sex of the chick. Thus, the pose management system 230 can determine to use or select the image or data associated with the image. As another example, the pose management system 230 can determine that a sex determination associated with the image should be given a higher weight value or confidence parameter than a sex determination associated with an unsatisfactory image.
[0061] As described herein, the pose management system 230 can receive one or more images (e.g., a video stream). The pose management system 230 can select at least one image from the plurality of images. In some cases, the pose management system 230 can select an image according to an image selection strategy. For example, the image selection strategy can indicate selecting the "best" image, where the best image can be defined as the image by which the most accurate or reliable sex prediction can be determined. As another example, the image selection strategy can indicate selecting an image that meets the pose criteria as described herein. Additionally, as another example, the image selection strategy can indicate selecting an image corresponding to a particular time or a particular pose (e.g., the presence of outstretched wings). If there is no image, e.g., if multiple images do not meet the pose criteria, the image selection strategy can indicate not selecting any image.
[0062] The sex indicator 240 can determine the sex of the chick and / or a confidence parameter associated with the sex determination. In some cases, the sex indicator 240 can determine multiple sex indications for a single chick. For example, as described, the imaging system 220 can obtain multiple images of a chick moving along the path of the path system 210. In some such cases, the sex indicator 240 can generate a sex determination for one, multiple, or each of the multiple images. Additionally, in some cases, the sex indicator 240 can generate a sex determination for one or both wings of the chick. As a non-limiting example, in some cases, the right wing of the chick can indicate that the chick is male with a 75% confidence, while the left wing of the chick can indicate that the chick is male with an 82% confidence.
[0063] Using any one of a variety of techniques, the sex indicator 240 can determine the sex of the chick. For example, in some cases, the sex indicator 240 implements a feather detection technique to determine the presence of coverts and primary flight feathers on the wings. Additionally, the sex indicator 240 can evaluate one or more sex-related characteristics of the coverts and primary flight feathers to determine whether the chick is male or female. Sex-related characteristics can include, but are not limited to, the length of one or more primary flight feathers, the length of one or more coverts, or the relationship between one or more primary flight feathers and one or more coverts. The relationship can include a relative size or length comparison, such as which of the primary flight feathers or coverts is longer. For example, as described herein, if the primary flight feathers are longer than the coverts, the chick is female; if the primary flight feathers and coverts are the same length or the primary flight feathers are shorter than the coverts, the chick is male.
[0064] In some embodiments, one or more portions of the sex determination can involve any one or more of a variety of machine learning techniques, such as supervised machine learning (e.g., decision trees, nearest neighbors, support vector machines, neural networks, naive Bayes classifiers, etc.) and / or unsupervised machine learning (e.g., clustering, principal component analysis, etc.). As a non-limiting example, a supervised machine learning method can be employed to train a neural network to identify and classify coverts and primary flight feathers. In some such cases, the machine learning algorithm can include one or more layers for making decisions based on target features on the chick seen in the image. Each layer evaluates the image and makes a decision based on previously trained data and feeds that decision back to the main algorithm. Target features include the body position of the chick, the body orientation, the presence of outstretched wings, the presence / identification of coverts and primary flight feathers on the wings, and the characteristics of the coverts and primary flight feathers that indicate sex. Decisions regarding these features will be used together to evaluate the sex of the chick and the quality / certainty of that decision. If multiple images include reliable data regarding the sex of the chick, the algorithm combines the data from multiple images to make a final decision on the sex of the chick.
[0065] Based on the sex determination, the classification system 250 can assist in classifying the chicks. In some cases, the classification system 250 includes an automatic classification device that automatically classifies the chicks based on the sex determination. For example, the classification device can automatically classify or direct the chicks to desired locations, such as bins or coops for a particular sex. Additionally or alternatively, the chicks can be manually classified by staff based on the sex determination.
[0066] Figure 2B - 2FShows an example comparison of the wing feathers of male and female chicks. Specifically, FIGS. 2B and 2C show example wings of female chicks, where the coverts 204 are shorter than the primaries 202. Additionally, FIGS. 2D-2F show example wings of male chicks, where the coverts 204 are the same length as the primaries 202 (see FIG. 2D) or longer than the primaries 202 (see FIGS. 2E and 2F).
[0067] Chick sex determination is a method of differentiating the sex of chickens or other young chicks, such as to separate female chicks or "pullets" from male chicks or "cockerels". Feather sex determination is a method of chick sex determination based on the growth rate of the wing feathers of a chick. Specifically, if the primary flight feathers of a chick are longer than its coverts, the chick is identified as female; if the coverts and primary flight feathers are the same length or the primary flight feathers are shorter than the coverts, the chick is identified as male. As described herein, manual feather sex determination (e.g., where a trained worker physically spreads the wings of a chick by hand) can damage the wings of the chick. Additionally, manual feather sex determination is generally unreliable at scale, e.g., due to visual fatigue or other issues associated with human error.
[0068] To address these or other challenges, the avian sex determination system 200 can automatically and accurately determine the sex of a chick by obtaining and processing one or more images of the chick to identify sex-related features from the chick's wings and make a sex determination. Additionally, since feather sex determination involves examining the outstretched wings of a chick, the avian sex determination system 200 can implement techniques to cause the chick to reflexively extend its wings, thereby more favorably reducing the likelihood of injury to the chick during feather sex determination compared to manually spreading the wings of the chick by hand. Additionally, the avian sex determination system 200 advantageously increases the speed / productivity at which sex is determined, improves the accuracy of feather sex determination by reducing errors associated with human intervention, and facilitates large-scale chick sex determination.
[0069] Figure 3 Is a flowchart of an exemplary procedure 300 for determining the sex of a chick. Although described as being implemented by the avian sex determination system 200, it should be understood that the elements outlined for procedure 300 can be implemented by any one or any combination of physical structures, hardware components, or computing devices associated with the avian sex determination system 200. Accordingly, the following exemplary embodiments should not be construed as limiting.
[0070] At block 302, the avian sex determination system 200 causes the chick to move along a predetermined path. As described herein, the avian sex determination system 200 can define a path along which the chick can move as part of the sex determination process. In some cases, the path includes a physical structure that defines a generally pre-planned path or trajectory. For example, the path can include a chute or other negatively sloped route, a passageway, a tunnel, etc., along which the chick can navigate or be navigated (e.g., by a cart, a conveyor belt, etc.).
[0071] As described herein, the path can be associated with one or more stimuli designed to encourage the chick to transition into (or remain in) an open posture. The implementation of the stimuli can vary by embodiment. For example, in some cases, the stimuli can refer to the slope or angle of the path, the speed, position, or direction of the chick as it moves along the path, or external actions or measures that affect the chick as it moves along the path (e.g., noise, air flow, or water flow). In some such cases, the chick moving along the path associated with the stimuli can cause the chick to feel a sensation (e.g., a sense of falling similar to what a human might feel), which prompts the chick to reflexively open or spread its wings (e.g., to maintain balance). Thus, one or more stimuli can encourage, motivate, or trigger the chick to transition into or remain in an open posture. It will be appreciated that the stimuli can be passively associated with the path (e.g., the curve, slope, or transition of the path). Additionally or alternatively, the stimuli can be applied or controlled by the avian sex determination system 200 (e.g., the speed at which the chick moves along the path, the vibration of the conveyor belt, the presence / absence of external actions, etc.).
[0072] At block 304, the avian sex determination system 200 obtains image data including at least one image of the chick. In some cases, the at least one image includes an image of the chick before and / or after moving along the path. In some cases, the at least one image includes an image of the chick as it moves along the path. For example, the at least one image can include an image of the chick near the stimuli (e.g., at or shortly after the stimuli) to increase the likelihood of capturing an image of the chick in an open posture. As another example, the at least one image can include a series of images of the chick as it moves along the path (e.g., X frames per second), which can facilitate increasing the likelihood that the at least one image depicts the chick in an open posture. In some cases, the avian sex determination system 200 obtains the image data in real time. For example, the avian sex determination system 200 can receive and process images as the chick moves along the path. In some cases, the avian sex determination system 200 obtains the image data from a data storage. For example, the image data can be stored in a local or remote data storage, from which the avian sex determination system 200 can retrieve the image data.
[0073] At block 306, based on an image selection strategy, the avian sex determination system 200 selects at least one image from among the plurality of images. As described herein, in some cases, chick images may not be helpful in determining the sex of a chick. For example, sex-related features of a chick may be hidden in the image. Thus, in some cases, as part of a filtering process, based on an image selection strategy, the avian sex determination system 200 may identify and / or select one, some, or all of the plurality of images. For example, the image selection strategy may indicate selection of the "best" image, where the best image may be defined as the image from which the most accurate or reliable sex prediction can be determined. As another example, the image selection strategy may indicate selection of images that meet a particular pose criterion, as described herein. Additionally, as another example, the image selection strategy may indicate selection of images corresponding to a particular time or a particular pose (such as the presence of outstretched wings). In some cases, if there are no images (such as if none of the plurality of images meet the pose criterion), the image selection strategy may indicate not selecting any images.
[0074] In some cases, to select at least one image, the avian sex determination system 200 processes the images (e.g., in real time or near real time) to determine whether the image includes a chick and / or whether the pose of the chick meets the pose criterion. In this way, the avian sex determination system 200 can evaluate the image to determine whether the image is likely to be used to accurately determine the sex of the chick. The avian sex determination system 200 can evaluate the image to determine whether the image is satisfactory. In some cases, at least partially based on a determination that the image is likely to be used to accurately determine the sex of the chick or a determination that the chick in the image meets the pose criterion, the avian sex determination system 200 determines that the image is satisfactory. Inevitably, at least partially based on a determination that the image is not likely to be used to accurately determine the sex of the chick or a determination that the chick in the image does not meet the pose criterion, the avian sex determination system 200 may determine that the image is not satisfactory. In some cases, selecting at least one image includes selecting one, some, or all of the images identified as satisfactory.
[0075] At block 308, the avian sex determination system 200 identifies one or more sex-related features of the chick in the selected at least one image. Sex-related features may include, but are not limited to, the length of one or more primary flight feathers, the length of one or more coverts, or the relationship between one or more primary flight feathers and one or more coverts. The relationship may include a relative size or length comparison, such as which of the primary or covert feathers is longer.
[0076] At block 310, the avian sex determination system 200 determines the sex, at least in part, based on one or more sex-related characteristics. As described herein, if the primary flight feathers of a chick are longer than the coverts, the chick is determined to be female; if the coverts and primary flight feathers are of the same length or the primary flight feathers are shorter than the coverts, the chick is determined to be male. As described herein, in some cases, the avian sex determination system 200 can determine a sex identification indication associated with each wing including visible sex-related characteristics.
[0077] At block 312, the avian sex determination system 200 determines a confidence parameter associated with the sex determination at block 310. In some cases, the confidence parameter is a number between 0 and 1, indicating the likelihood that the output of the machine learning model is correct and meets the user's request. For example, the output of the avian sex determination system 200 (e.g., sex determination) can consist of one or more predictions. In some such cases, each prediction can be assigned a confidence score, and the higher the score, the greater the confidence of the avian sex determination system 200 in the correctness of the sex determination. In some embodiments, a confidence score higher than 0.7 indicates that the prediction is likely to be very accurate, while a score lower than 0.3 may indicate an uncertain prediction.
[0078] At block 314, the chicks are classified according to the sex determination and / or based on the confidence parameter. In some cases, the avian sex determination system 200 includes an automatic classification device. For example, the classification device can automatically classify or direct the chicks to desired locations, such as bins or coops of a particular sex. As another example, the avian sex determination system 200 may not include an automatic classification device. Instead, the sex determination and / or confidence parameter can be determined, and the avian sex determination system 200 can output an indication of the determination (e.g., visual display, audible noise, etc.). In response, the staff can manually classify the chicks according to the indication provided by the avian sex determination system 200. In some cases, for example, when the confidence parameter indicates an uncertain sex, the chicks can be classified into a separate bin corresponding to the "unknown" sex and / or the program 300 can be rerun to re-evaluate the chicks.
[0079] Fewer, more, or different blocks may be used as part of program 300. In some cases, one or more blocks may be omitted. In some embodiments, the blocks of program 300 may be combined with any one or any combination of other blocks of program 300. Additionally, for the same or different chicks, program 300 may be executed multiple times, such as dozens, hundreds, or thousands of times. For example, the avian sex determination system 200 may execute program 300 for each chick in a group of hundreds or thousands of chicks. In some such cases, program 300 may be executed simultaneously or sequentially. For example, in some cases, the avian sex determination system 200 executes program 300 for one chick at a time, sequentially for each chick. Alternatively, the avian sex determination system 200 may execute program 300 for two or more chicks simultaneously. For example, the avian sex determination system 200 may include multiple predetermined paths for one or more chicks to travel simultaneously, or the avian sex determination system 200 may include one predetermined path for multiple chicks to travel simultaneously. In some such cases (e.g., where the image includes multiple chicks), the avian sex determination system 200 may perform additional image processing to identify and / or track a particular chick in the image.
[0080] In some cases, the avian sex determination system 200 may execute program 300 or a portion of program 300 for each of multiple images of a chick moving along a path. For example, the avian sex determination system 200 may obtain images in real time or near real time, and may perform one or more steps of program 300 on the obtained images before receiving any other images or independently of receiving any other images.
[0081] Figure 4A - 4F is an example post - processing top view of chick 404 moving along example path 402. In these example figures, path 402 includes a chute and chick 404 moves downward along path 402 in a backward direction, successively moving between Figure 4A - 4F the images shown. Figure 4A - 4F is an example of the output of the avian sex determination system 200 as part of executing program 300. In these examples, the primary flight feathers of the chick are longer than its coverts, so the chick is identified as female.
[0082] Figure 4A depicts chick 404 at a first time as chick 404 moves along path 402. For this image 410, the avian sex determination system 200 determines that neither the left wing 406L nor the right wing 406R of the chick is in a satisfactory pose for sex determination. Therefore, the avian sex determination system 200 does not Figure 4A generate a sex judgment.
[0083] Figure 4BDepicts chick 404 at a second time as it moves along path 402. For this image 420, the avian sex determination system 200 determines that both the left wing 406L and the right wing 406R of the chick are in satisfactory poses for sex determination. For the left wing 406L, the avian sex determination system 200 determines that the sex-related feature 408L indicates that the chick is female. Thus, based on the sex-related feature 408L of the left wing 406L, the avian sex determination system 200 outputs a sex assignment 410L of "female". Additionally, the avian sex determination system 200 determines that the confidence parameter 412L associated with the sex assignment 410L is 0.93. For the right wing 406R, the avian sex determination system 200 determines that the sex-related feature 408R indicates that the chick is female. Thus, based on the sex-related feature 408R of the right wing 406R, the avian sex determination system 200 outputs a sex determination 410R of "female". Additionally, the avian sex determination system 200 determines that the confidence parameter 412R associated with the sex determination 410R is 0.90.
[0084] Figure 4C Depicts chick 404 at a third time as it moves along path 402. For this image 430, the avian sex determination system 200 determines that neither the left wing 406L nor the right wing 406R of the chick is in a satisfactory pose for sex determination. Thus, the avian sex determination system 200 does not Figure 4C generate a sex determination.
[0085] Figure 4D Depicts chick 404 at a fourth time as it moves along path 402. For this image 440, the avian sex determination system 200 determines that the left wing 406L of the chick is in a satisfactory pose for sex determination. However, the avian sex determination system 200 determines that the right wing 406R is not in a satisfactory pose for sex determination. Thus, the avian sex determination system 200 does not generate a sex determination for the right wing 406R. For the left wing 406L, the avian sex determination system 200 determines that the sex-related feature 408L indicates that the chick is female. Thus, based on the sex-related feature 408L of the left wing 406L, the avian sex determination system 200 outputs a sex determination 410L of "female". Additionally, the avian sex determination system 200 determines that the confidence parameter 412L associated with the sex determination 410L is 0.91.
[0086] Figure 4E Depicts chick 404 at a fifth time as it moves along path 402. For image 450, the avian sex determination system 200 determines that neither the left wing 406L nor the right wing 406R of the chick is in a satisfactory pose for sex determination. Thus, the avian sex determination system 200 does notFigure 4E Generate gender determination.
[0087] Figure 4F Depicts chick 404 moving along path 402 at a sixth time. For this image 460, the avian gender identification system 200 determines that both the left wing 406L and the right wing 406R of the chick are in satisfactory poses for determining gender. For the left wing 406L, the avian gender identification system 200 determines that the gender-related feature 408L indicates that the chick is female. Thus, based on the gender-related feature 408L of the left wing 406L, the avian gender identification system 200 outputs a gender determination 410L of "female". Additionally, the avian gender identification system 200 determines that the confidence parameter 412L associated with the gender determination 410L is 0.93. For the right wing 406R, the avian gender identification system 200 determines that the gender-related feature 408R indicates that the chick is female. Thus, based on the gender-related feature 408R of the right wing 406R, the avian gender identification system 200 outputs a gender determination 410R of "female". Additionally, the avian gender identification system 200 determines that the confidence parameter 412R associated with the gender determination 410R is 0.91.
[0088] As described herein, in some cases, the avian gender identification system 200 can use only one image (e.g., Figure 4B , 4D and any one of images 420, 440, 460 of 4F) to determine the gender of chick 404. Alternatively, in some cases, the avian gender identification system 200 can use data from a combination of two or more images to determine the gender of chick 404. For example, based on the associated confidence parameters, the avian gender identification system 200 can weight two or more determinations. In this case, for example, since images 420, 440, 460 all indicate that the chick is female, the avian gender identification system 200 can output an overall gender determination of "female" and can accordingly classify the chick.
[0089] Figure 5A - 4F is an example post-processing example top view of chick 504 moving along example path 502. In these example figures, path 502 includes a chute and chick 504 moves downward along path 502 in a backward direction, successively moving between the Figure 5A - 4F shown images. Figure 5A - 4F is an example of the output of the avian gender identification system 200 as part of executing program 300. In these examples, the primary flight feathers of the chick are shorter than its coverts, and thus the chick is identified as male.
[0090] Figure 5ADepicts chick 504 at a first time as it moves along path 502. For this image 510, the avian sex determination system 200 determines that both the left wing 506L and the right wing 506R of the chick are in satisfactory poses for sex determination. For the left wing 506L, the avian sex determination system 200 determines that the sex-related feature 508L indicates that the chick is male. Thus, based on the sex-related feature 508L of the left wing 506L, the avian sex determination system 200 outputs a sex determination 510L of "male". Additionally, the avian sex determination system 200 determines that the confidence parameter 512L associated with the sex determination 510L is 0.90. For the right wing 506R, the avian sex determination system 200 determines that the sex-related feature 508R indicates that the chick is male. Thus, based on the sex-related feature 508R of the right wing 506R, the avian sex determination system 200 outputs a sex determination 510R of "male". Additionally, the avian sex determination system 200 determines that the confidence parameter 512R associated with the sex determination 510R is 0.73.
[0091] Figure 5B Depicts chick 504 at a second time as it moves along path 502. For this image 520, the avian sex determination system 200 determines that the right wing 506R of the chick is in a satisfactory pose for sex determination. However, the avian sex determination system 200 determines that the left wing 506L is not in a satisfactory pose for sex determination. Thus, the avian sex determination system 200 does not generate a sex judgment for the left wing 506L. For the right wing 506R, the avian sex determination system 200 determines that the sex-related feature 508R indicates that the chick is male. Thus, based on the sex-related feature 508R of the right wing 506R, the avian sex determination system 200 outputs a sex determination 510R of "male". Additionally, the avian sex determination system 200 determines that the confidence parameter 512R associated with the sex determination 510R is 0.87.
[0092] Figure 5CDepicts chick 504 at a third time as chick 504 moves along path 502. For this image 530, the avian sex determination system 200 determines that both the left wing 506L and the right wing 506R of the chick are in satisfactory poses for sex determination. For the left wing 506L, the avian sex determination system 200 determines that the sex-related feature 508L indicates that the chick is male. Accordingly, based on the sex-related feature 508L of the left wing 506L, the avian sex determination system 200 outputs a sex determination 510L of "male". In addition, the avian sex determination system 200 determines that the confidence parameter 512L associated with the sex determination 510L is 0.90. For the right wing 506R, the avian sex determination system 200 determines that the sex-related feature 508R indicates that the chick is male. Accordingly, based on the sex-related feature 508R of the right wing 506R, the avian sex determination system 200 outputs a sex determination 510R of "male". In addition, the avian sex determination system 200 determines that the confidence parameter 512R associated with the sex determination 510R is 0.92.
[0093] Figure 5D Depicts chick 504 at a fourth time as chick 504 moves along path 502. For this image 540, the avian sex determination system 200 determines that both the left wing 506L and the right wing 506R of the chick are in satisfactory poses for sex determination. For the left wing 506L, the avian sex determination system 200 determines that the sex-related feature 508L indicates that the chick is male. Accordingly, based on the sex-related feature 508L of the left wing 506L, the avian sex determination system 200 outputs a sex determination 510L of "male". In addition, the avian sex determination system 200 determines that the confidence parameter 512L associated with the sex determination 510L is 0.88. For the right wing 506R, the avian sex determination system 200 determines that the sex-related feature 508R indicates that the chick is male. Accordingly, based on the sex-related feature 508R of the right wing 506R, the avian sex determination system 200 outputs a sex determination 510R of "male". In addition, the avian sex determination system 200 determines that the confidence parameter 512R associated with the sex determination 510R is 0.89.
[0094] Figure 5EDepicts chick 504 at a fifth time as it moves along path 502. For this image 550, the avian sex determination system 200 determines that both the left wing 506L and the right wing 506R of the chick are in satisfactory poses for sex determination. For the left wing 506L, the avian sex determination system 200 determines that the sex-related feature 508L indicates that the chick is male. Accordingly, based on the sex-related feature 508L of the left wing 506L, the avian sex determination system 200 outputs a sex determination 510L of "male". In addition, the avian sex determination system 200 determines that the confidence parameter 512L associated with the sex determination 510L is 0.82. For the right wing 506R, the avian sex determination system 200 determines that the sex-related feature 508R indicates that the chick is male. Accordingly, based on the sex-related feature 508R of the right wing 506R, the avian sex determination system 200 outputs a sex determination 510R of "male". In addition, the avian sex determination system 200 determines that the confidence parameter 512R associated with the sex determination 510R is 0.90.
[0095] Figure 5F Depicts chick 504 at a sixth time as it moves along path 502. For this image 560, the avian sex determination system 200 determines that both the left wing 506L and the right wing 506R of the chick are in satisfactory poses for sex determination. For the left wing 506L, the avian sex determination system 200 determines that the sex-related feature 508L indicates that the chick is male. Accordingly, based on the sex-related feature 508L of the left wing 506L, the avian sex determination system 200 outputs a sex determination 510L of "male". In addition, the avian sex determination system 200 determines that the confidence parameter 512L associated with the sex determination 510L is 0.92. For the right wing 506R, the avian sex determination system 200 determines that the sex-related feature 508R indicates that the chick is male. Accordingly, based on the sex-related feature 508R of the right wing 506R, the avian sex determination system 200 outputs a sex determination 510R of "male". In addition, the avian sex determination system 200 determines that the confidence parameter 512R associated with the sex determination 510R is 0.73.
[0096] As described herein, in some cases, the avian sex determination system 200 can determine the sex of chick 504 using only one image. Alternatively, in some cases, the avian sex determination system 200 can use combined data from two or more images to determine the sex of chick 504. For example, based on the associated confidence parameters, the avian sex determination system 200 can weight two or more judgments. In this case, for example, since each image indicates that the chick is male, the avian sex determination system 200 can output an overall sex judgment of "male" and can classify the chick accordingly.
[0097] Figure 6 Displays a bar graph 500 illustrating an example relationship between the age of chicks and the accuracy rate of sex determination using the techniques and / or systems described herein. The X-axis classifies the flock of chickens into three age groups: old, prime, and young. The Y-axis quantifies the accuracy rate of sex determination as a percentage.
[0098] For the age group labeled "old" (including birds with an age range greater than or equal to 58 days), the bar graph indicates an accuracy rate of sex determination of 96.3%. The age group labeled "prime" (which includes birds between 35 and 58 days old) indicates an accuracy rate of 98.5%. The age group classified as "young" (representing birds between 0 and 30 days old) indicates an accuracy rate of sex determination of 98.0%.
[0099] Figure 6 The data shown in is from two validation tests conducted on the system. The first test involved 61,800 bird samples, while the second test covered 80,000 birds. Figure 6 The percentage accuracy rates shown in are based on the combined results of these two tests.
[0100] It should be recognized that the age classifications or ranges used to define the "young", "prime", and "old" groups vary between different embodiments of the system. For example, in various embodiments, "young" birds can be those whose age falls within less than 10, 15, 20, 25, 30, 35, or 40 days, with a possible variation of a few days. Similarly, "old" birds can be defined as those whose age is 30, 35, 40, 45, 50, 55, 60, 65, 70, or 75 days or older, also allowing for a variation of a few days. Birds classified as "prime" generally fall within the age range between the "young" and "old" classifications. In some cases, there may be a buffer period of several days for differentiating between age groups, such as between "young" and "prime" or between "prime" and "old".
[0101] Embodiments of the present invention provide systems, methods, and computer-readable media with instructions for identifying the sex of chicks. A method includes moving a chick along a path of a stimulus associated with eliciting an open posture of the chick. The method further includes obtaining one or more images of the chick in the open posture, determining one or more sex-related characteristics of the chick at least in part based on the one or more images, and determining the sex of the chick at least in part based on the one or more sex-related characteristics. One or more parts of the method can include any of various machine learning techniques, such as supervised machine learning and / or unsupervised machine learning. For example, a supervised machine learning method can be employed to train a neural network to recognize the posture of the chick, identify sex-related characteristics associated with the wings of the chick, and / or evaluate sex-related characteristics for sex determination.
[0102] Term Any and / or all of the above features and functions may be combined with each other, unless otherwise stated above or any such embodiments may be incompatible due to their functions or structures, as will be apparent to those of ordinary skill in the art. Unless contrary to physical possibility, it is contemplated that the methods / steps described herein may be performed in any order and / or in any combination, and the components of the various embodiments may be combined in any manner.
[0103] Although the subject matter has been described in language specific to structural features and / or acts, it is to be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or acts described above. Rather, the above specific features and acts are disclosed as examples for implementing the claims, and other equivalent features and acts are intended to be within the scope of the claims.
[0104] Unless otherwise expressly stated or understood in context, conditional language such as "able to", "can", "may", or "might" generally is intended to convey that certain embodiments include certain features, elements, and / or steps, while other embodiments do not include those features, elements, and / or steps. Thus, such conditional language generally is not intended to imply that the features, elements, and / or steps are necessary for one or more embodiments or that one or more embodiments necessarily include logic for deciding whether these features, elements, and / or steps are included in any particular embodiment or are to be performed in any particular embodiment (whether or not user input or prompting is required).
[0105] Unless the context clearly requires otherwise, throughout the specification and claims, the words "comprising", "including", and the like are to be construed in an inclusive sense as opposed to an exclusive or exhaustive sense; for example, in the sense of "including, but not limited to". The terms "connected", "coupled", or any variant thereof as used herein denote any direct or indirect connection or coupling between two or more elements; the coupling or connection between elements may be physical, logical, or a combination of both. Further, the words "herein", "above", "below", and words of similar import when used in this application refer to the entire application and not to any particular part of this application. Where context permits, words using the singular or plural number may also respectively include the plural or singular number. When referring to a list of two or more items, the word "or" covers all of the following interpretations: any of the items in the list, all of the items in the list, and any combination of the items in the list. Similarly, when referring to a list of two or more items, the term "and / or" covers all of the following interpretations: any of the items in the list, all of the items in the list, and any combination of the items in the list.
[0106] Unless otherwise expressly stated, the conjunctive term such as the phrase "at least one of X, Y, and Z" shall be understood in context as typically used to express that an item, term, etc. can be X, Y, or Z or any combination thereof. Thus, such conjunctive terms are generally not intended to mean that certain embodiments require the separate presence of at least one X, at least one Y, and at least one Z. Further, the phrase "at least one of X, Y, or Z" is typically used to express that an item, term, etc. can be X, Y, or Z or any combination thereof.
[0107] Degree terms used herein, such as the terms "about", "approximately", "generally", and "substantially" used herein, represent a value, quantity, or property that is close to the stated value, quantity, or property but still enables the required function or achieves the required result. For example, the terms "about", "approximately", "generally", and "substantially" can refer to a quantity within less than 10%, less than 5%, less than 1%, less than 0.1%, and less than 0.01% of the stated quantity. As another example, in certain embodiments, the terms "substantially parallel" and "essentially parallel" refer to a value, quantity, or property that deviates from perfect parallelism by less than or equal to 10 degrees, 5 degrees, 3 degrees, or 1 degree. As another example, in certain embodiments, the terms "substantially perpendicular" and "essentially perpendicular" refer to a value, quantity, or property that deviates from perfect perpendicularity by less than or equal to 10 degrees, 5 degrees, 3 degrees, or 1 degree.
[0108] Any term typically associated with a circle, such as "radius", "radial", "diameter", "circumference", "circumferential", or any derivative or similar type of term, is intended to be used to denote any corresponding structure in any type of geometric structure, and not just circular structures. For example, "radial" applied to another geometric structure should be understood to refer to the direction or distance between a location corresponding to the approximate geometric center of such a structure and the perimeter of such a structure; "diameter" applied to other geometric structures should be understood to refer to the cross-sectional width of such a structure; "circumference" applied to other geometric structures should be understood to refer to the peripheral region. Nothing in this specification or the drawings should be construed to limit these terms to circular or circular structures.
[0109] All patents, applications, and other references mentioned above, including any that may be listed in the accompanying application files, are hereby incorporated by reference into this text. If necessary, various aspects of the inventive concept may be modified to adopt the systems, functions, and concepts of the various references described above, thereby providing further embodiments of the inventive concept. Given the above "Detailed Description", these and other changes may be made to the inventive concept. Although the above description describes certain examples of the inventive concept and describes the expected best mode, no matter how detailed the above description may seem in the text, the inventive concept may be practiced in many ways. The details of the system may vary greatly in its specific implementation, but are still included in the inventive concept disclosed herein. As mentioned above, the specific terms used when describing certain features or aspects of the inventive concept should not be construed as implying that the terms are redefined herein to be limited to any specific characteristics, features, or aspects of the inventive concept associated with the terms. Generally speaking, unless the above "Detailed Description" section clearly defines these terms, the terms used in the following claims should not be construed as limiting the inventive concept to the specific examples disclosed in the specification. Therefore, the actual scope of the inventive concept not only covers the disclosed examples, but also covers all equivalent ways of practicing or implementing the inventive concept in the claims.
[0110] To reduce the number of claims, certain aspects of the inventive concept are presented in certain claim forms, but the applicant contemplates presenting other aspects of the inventive concept in any number of claim forms. Any claim intended to be treated under 35 U.S.C. § 112(f) begins with "means for...", but the use of the term "means for" in any other context is not intended to invoke the treatment of 35 U.S.C. § 112(f). Accordingly, the applicant reserves the right to add claims to this application or a subsequent application after the filing of this application.
Claims
1. A computing system, comprising: a memory; and one or more processors, coupled to the memory and configured to: obtain image data comprising a set of images of chicks, wherein the set of images of chicks corresponds to discrete time periods during which the chicks move along a predetermined path, wherein the predetermined path is associated with a stimulus for eliciting an open posture of the chicks, wherein the open posture comprises at least partial extension of at least one wing of the chicks; determine one or more sex-related characteristics of the chicks based on the image data; and determine the sex of the chicks based on the one or more sex-related characteristics.
2. The system according to claim 1, wherein the determination of the sex of the chicks is achieved with an accuracy equal to or exceeding 96%.
3. The system according to claim 1, wherein the one or more processors are further configured to select an image from the set of images based on a pose selection strategy, and wherein the selected image is input into a trained neural network.
4. The system according to claim 3, wherein the pose selection strategy indicates selecting an image in which the pose of the chicks meets a pose criterion.
5. The system according to claim 4, wherein the one or more processors are further configured to determine that the pose of the chicks in the selected image meets the pose criterion.
6. The system according to any one of claims 3 or 4, wherein the pose criterion comprises at least one of a body position threshold, a body orientation threshold, a wing position threshold, or a threshold wing orientation threshold.
7. The system according to claim 3, wherein the pose selection strategy indicates selecting an image depicting the chicks with extended wings.
8. The system according to any one of claims 1 to 7, wherein the one or more processors are further configured to: for each image in the set of images: determine the corresponding pose of the chicks; and determine whether the corresponding pose meets the pose criterion.
9. The system according to any one of claims 1 to 8, wherein the set of images comprises a plurality of images, and wherein each image in the set of images corresponds to a different time of the discrete time periods.
10. The system according to any one of claims 1 to 9, wherein the one or more sex-related characteristics comprise the length of a set of primary flight feathers or the length of a set of coverts.
11. The system according to claim 10, wherein the length of the set of coverts is less than the length of the set of primary flight feathers, and wherein the sex output indicates that the chicks are female.
12. The system according to claim 10, wherein the length of the set of coverts is equal to or greater than the length of the set of primary flight feathers, and wherein the sex output indicates that the chicks are male.
13. The system according to any one of claims 1 to 12, wherein the predetermined path is a chute, wherein the stimulus is a downwardly inclined plane of the chute, and wherein moving downward along the chute causes the chicks to transition to the open posture.
14. The system according to any one of claims 1 to 12, wherein the predetermined path comprises a conveyor belt.
15. The system according to claim 14, wherein the stimulus is a change in speed, a descent, or a vibration of the conveyor belt.
16. The system according to any one of claims 1 to 15, wherein the stimulus comprises noise from a noise machine or a fluid spray activated by a fluid sprayer.
17. The system according to any one of claims 1 to 16, wherein the stimulation triggers the chick to reflexively enter an open posture.
18. The system according to any one of claims 1 to 17, wherein one or more processors are further configured to automatically classify the chicks according to the gender identification output.
19. The system according to any one of claims 1 to 18, wherein each image in a set of images is a real-time image.
20. A non-transitory computer-readable medium comprising computer-executable instructions that, when executed by a computing system of a data acquisition and query system, cause the computing system to: obtain image data comprising a set of images of chicks, wherein the set of images of chicks corresponds to discrete time periods during which the chicks move along a predetermined path, wherein the predetermined path is associated with a stimulation for eliciting an open posture of the chicks, and wherein the open posture comprises at least partial extension of at least one wing of the chicks; determine one or more gender-related characteristics of the chicks based on the image data; and determine the gender of the chicks based on the one or more gender-related characteristics.
21. A method for identifying the gender of chicks, the method comprising: obtaining a set of images of a chick, wherein the set of images corresponds to discrete time periods during which the chick moves along a predetermined path, wherein the predetermined path is associated with a stimulation for eliciting an open posture of the chick, and wherein the open posture comprises at least partial extension of at least one wing of the chick; inputting at least one image in the set of images of the chick into a trained neural network that characterizes one or more gender-related characteristics of the chick; obtaining a gender identification and confidence parameter output from the trained neural network, wherein the gender identification output indicates the likely gender of the chick, and wherein the confidence parameter output indicates the confidence associated with the gender identification output; and causing an action at least partially based on the gender identification and confidence parameter outputs.
Citation Information
Cited By
Chick gender detection system
CN120458039A
