System

Abstract

A weight determination system and method for determining a weight of an animal comprises an imaging device that obtains an image of the animal and an image processor that processes pixel information from the image to determine a weight of the animal, wherein the processed pixel information includes pixel values in the image. Further provided is a method for determining a weight of an animal which comprises the steps of collecting a set of data including pixel information in one or more images, wherein the pixel information includes pixel values in the or each image, creating a training set including the collected set of data, training a machine learning algorithm or model using the training set, and determining the weight of the animal based on an output of the machine learning algorithm or model. The animal may be a bird, such as a chicken or chick.

Description

This invention relates to a weight determination system, computer-implemented methods of determining a weight of an animal, and a computer program comprising computer code configured to perform the computer-implemented methods. Measurement of animal weight is used to monitor the growth and health of animals. According to a first aspect of the invention, there is provided a weight determination system for determining a weight of an animal, the weight determination system comprising: an imaging device for obtaining an image of the animal; and an image processor programmed to process pixel information in the image to determine a weight of the animal, wherein the processed pixel information includes pixel values in the image. The image processor may be or form part of a computing device. The computing device may be, may include or may form part of one or more of an electronic device, a portable electronic device, a portable telecommunications device, a mobile phone, a personal digital assistant, a tablet, a phablet, a laptop computer, a server, a cloud computing network, a smartphone, a smartwatch, smart eyewear, and a module for one or more of the same. In embodiments of the invention, the processed pixel information may include: pixel colour values in the image; pixel brightness values in the images; a pixel count in the image; a geometrical relationship between pixels in the image; and / or a pixel pattern in the image. In further embodiments of the invention, the image processor may be programmed to process red pixel information in the image to determine the weight of the animal. The image processor may be programmed to perform pixel segmentation on selected pixels (e.g. red pixels) in the image prior to processing the pixel information in the image to determine the weight of the animal. Preferably the pixel segmentation includes reducing the pixel value of non-selected pixels in the image. More preferably the pixel segmentation includes reducing the pixel value of non-selected pixels in the image to zero. The image processor may be programmed to process pixel information in the image to find an animal in the image. The image processor may be programmed to process pixel information in the image to identify a type, sex, breed, age and / or health condition of the animal in the image so as to determine the weight of the animal. The invention is applicable to a wide range of animals. In embodiments of the invention, the animal may be a bird, preferably poultry, bovine or swine. The image processor may be programmed to process pixel information in the image to identify an integument pattern, an integument coverage and / or an integument texture of the bird so as to determine the weight of the animal. The integument may include, but is not limited to, feathers, fur or hair. The image processor may be programmed to process pixel information in the image to identify a crest of the bird so as to determine the weight of the animal. Preferably the image processor is programmed to process pixel information in the image to identify a shape and / or size of a crest of the bird so as to determine the weight of the animal. The image processor is programmed to create a weight profile for a single animal or a group of animals. The image processor may be programmed to perform pre-processing on the image prior to processing the pixel information in the image. The pre-processing may include: resizing the image; cropping the image to a fixed aspect ratio; and / or zooming in or out of an image. The imaging device may be arranged to, in use, obtain a top-down image of the animal. According to a second aspect of the invention, there is provided a computer-implemented method of determining a weight of an animal, the method comprising the steps of: obtaining an image of the animal; and processing pixel information in the image to determine the weight of the animal, wherein the processed pixel information includes pixel values in the image. The features and advantages of the first aspect of the invention and its embodiments apply mutatis mutandis to the features and advantages of the second aspect of the invention and its embodiments. According to a third aspect of the invention, there is provided a computer-implemented method of determining a weight of an animal, the method comprising the steps of: collecting a set of data, wherein the collected set of data includes pixel information in one or more images, wherein the pixel information includes pixel values in the or each image; creating a training set including the collected set of data; training a machine learning algorithm or model using the training set; and determining the weight of the animal based on an output of the machine learning algorithm or model. In embodiments of the invention, the machine learning algorithm or model may include, but is not limited to, an artificial neural network (such as a convolutional neural network). The features and advantages of the first and second aspects of the invention and their embodiments apply mutatis mutandis to the features and advantages of the third aspect of the invention and its embodiments. According to a fourth aspect of the invention, there is provided a computer program comprising computer code configured to perform the computer-implemented method of any one of the second and third aspects of the invention and their embodiments. The features and advantages of the first, second and third aspects of the invention and their embodiments apply mutatis mutandis to the features and advantages of the fourth aspect of the invention and its embodiments. It will be appreciated that the use of the terms "first" and "second", and the like, in this patent specification is merely intended to help distinguish between similar features, and is not intended to indicate the relative importance of one feature over another feature, unless otherwise specified. Within the scope of this application it is expressly intended that the various aspects, embodiments, examples and alternatives set out in the preceding paragraphs, and the claims and / or the following description and drawings, and in particular the individual features thereof, may be taken independently or in any combination. That is, all embodiments and all features of any embodiment can be combined in any way and / or combination, unless such features are incompatible. The applicant reserves the right to change any originally filed claim or file any new claim accordingly, including the right to amend any originally filed claim to depend from and / or incorporate any feature of any other claim although not originally claimed in that manner. Preferred embodiments of the invention will now be described, by way of non-limiting examples, with reference to the accompanying drawings in which: Figure 1 shows a weight determination system according to an embodiment of the invention; Figure 2 shows exploded and assembled views of a camera mount; Figure 3 shows a primary communications device; Figure 4 shows a secondary communications device; Figure 5 shows a communications network comprising primary and secondary communications devices; Figures 6 to 11 compare original images and pre-processed images of chickens; Figures 12 to 17 compare images before and after red pixel segmentation; Figures 18 to 22 shows various images of chickens recorded by the weight determination system of the invention; Figures 24 to 26 show pre-processed images of chickens with background masking; and Figures 27 to 29 show pre-processed images of chickens with segmented pixels. The figures are not necessarily to scale, and certain features and certain views of the figures may be shown exaggerated in scale or in schematic form in the interests of clarity and conciseness. The following embodiments of the invention are described with reference to the weight determination of poultry, particularly chicken. It will be appreciated that the following embodiments of the invention apply mutatis mutandis to other types of birds and animals. The poultry industry, and more specifically that of broiler chicken, is the world's largest meat protein category globally. As birds are highly sensitive, monitoring their growth rate is critical to optimising production, reducing food waste, and improving poultry health and welfare in this industry. Conventionally weight measurement of chickens is performed manually. A stock person would weigh chickens individually by hand on a regular basis. The weight data would be recorded onto paper and later converted into electronic data. Due to the high number of chickens per shed and the relatively short life cycle of the chickens, only a small percentage of the chickens are measured a small number of times in their life cycle, which results in weight data that is inaccurate and not representative of the chicken population. Furthermore, the manual weighing not only disrupts the natural behaviour of the chickens which can be stressful and detrimental to the health and welfare of the chickens, but also increases biosecurity risk. It is known to use an automatic weighing plate to measure a weight of a chicken that steps onto it. The automatic weighing plate is calibrated for every chicken flock and is configured to collect weights within a fixed tolerance range of the target weight for the chickens. The challenges with this approach are that the number of weight measurements reduces as the chickens age, which reduces the accuracy of the weight data, and that the collected weights are not a true reflection of the actual weights due to the use of a fixed tolerance range. The weight determination system of the invention enables accurate weight measurement of chickens that provides automated scalable accuracy and consistent profiling: • Automated scalable accuracy The weight determination system provides accurate measurements that are comparable to traditional weight scales and are improvable with more weight data. The scalability not only stems from the ability of the weight determination system to deliver the same performance from small farms (e.g. a few chickens) to very large farms (e.g. commercial sheds with more than 30,000 chickens) by just adding more equipment to cover the imaging region, but also stems from its automation that provides scalable accuracy for end users by just installing the weight determination system. • Consistent profiling The weight determination system provides consistency of measurements which facilitates the automated measurement of weight profiles over all chickens observed by the weight determination system. The weight profile is a key metric in the target industry, where the "shape" of the weight profile curve can be of more importance than the absolute weight values, for end users to understand the uniformity of weights across the chicken population. Consistent profiling by the weight determination system provides end users with valuable information, even if absolute weight values are biased for individual measurements. Other benefits of the invention include: • Improving use of finite labour resources by providing automatic measurements and more detailed representative data. • Eliminating bias associated with how humans and chickens interact with current measurement devices, such as the automatic weighing plate, so that better informed decisions can be made, slaughter planning becomes more accurate, and food waste can be reduced. • Reducing chicken and worker stress by eliminating and / or significantly reducing the need for manual handling of the chickens. • Decreasing biosecurity risk by accessing shed only when required and by reducing the amount of additional equipment (such as weighing scales) moved into and out of the shed. • Improving representativeness of data through the volume which is collected. • Enabling preventative measures to be undertaken through accurate measurements. A weight determination system according to an embodiment of the invention is shown in Figure 1. The weight determination system comprises an imaging device 30 and a computing device 32. The imaging device 30 is a camera for recording still images, although it may be a camera for recording moving images. The camera 30 may be a standard off-the-shelf camera, such as an RGB camera. The camera 30 is arranged inside a shed (or another type of enclosure) to face downwards towards the shed's floor in order to obtain top-down images of chickens inside the shed. It will be appreciated that the weight determination apparatus may use a single camera 30 or multiple cameras 30 in the same shed or across multiple sheds. It will be appreciated that the obtained top-down image may be of a single chicken or multiple chickens. The camera 30 may be configured to be protected against ingress of dust and airborne parties and may be designed to be waterproof. For example, the camera 30 may be designed to have an IP67 rating so that it is protected against dust inside the shed and can be left in place during washing of the shed. Figure 2 shows exploded and assembled views of an exemplary camera mount for holding the camera 30 in the downwards-facing arrangement. The camera mount comprises a swivel bracket 34, a rod 36 and a disc mount 38. The length of the rod 36 may vary depending on the desired height of the camera 30 from the shed's floor. In other embodiments, the rod 36 may be telescopic to permit adjustment of its length. One side of the disc mount 38 is attached to the camera 30, and the other side of the disc mount 38 is attached to a first end of the rod 36. The second end of the rod 36 is attached to the swivel bracket 34, which is attached to a ceiling of the shed. Alternatively, the swivel bracket 34 may be attached to a different overhead structure. The swivel bracket 34 allows adjustment of the camera's position and orientation to ensure that the camera 30 is facing downwards towards the shed's floor, even if the ceiling is sloped or slanted. One or more spirit levels 40 may be mounted onto the camera 30 to aid the adjustment of the camera's position and orientation to face downwards. The computing device 32 includes a processor and memory including computer program code. The memory and computer program code are configured to, with the processor, enable the computing device 32 to carry out various processing functions, such as an image processor carrying out an image processing function. It will be appreciated that references to a memory or a processor may encompass a plurality of memories or processors. In the embodiment shown, the computing device 32 is located in a remote location away from the shed. Image data is sent from the camera 30 to the computing device 32 via a cloud computing network 42. The image data is then processed by the computing device 32, and the processing results is saved in a data storage medium 44 which may or may not form part of the computing device 32. To facilitate exchange of image data between the camera 30 and the computing device 32, the weight determination system includes a primary communications device 46 that is installed inside or outside the shed. Figure 3 shows the primary communications device 46. The primary communications device 46 includes a communications controller 48 in the form of, e.g., a single board computer (SBC), which controls the transmission of image data from the camera 30 to the computing device 32 via the cloud computing network 42. The communications controller is in communication with the cloud computing network 42 via an internet router 50. The communications controller 48 may include a data storage medium, such as a flash drive, for temporarily storing image data in the event of a disruption in connectivity between the camera 30 and computing device 32, e.g. due to internet outage. Once the connection between the camera 30 and computing device 32 is restored, the stored image data is sent to the computing device 32. The communications controller 48 is housed inside an enclosure 52, which is preferably waterproof, e.g. with a IP65 rating. The enclosure 52 further houses a network switch 54, a power supply unit 56 connected to an external power supply 58, and a light indicator 62 (e.g. a light emitting diode, LED)showing the network status of the primary communications device. No light indicates that the primary communications device 46 has not yet powered up, a flashing light indicates that there is no network connection, and a solid non-flashing light indicates that there is a network connection. In a single shed setup with multiple cameras 30, all cameras 30 are in communication with the primary communications device 46. In a multi-shed setup with a single camera 30 or multiple cameras 30 per shed, all cameras 30 across all sheds are in communication with the primary communications device 46 via one or more secondary communications devices 64. The weight determination system includes multiple secondary communications devices 64, one per shed. Figure 4 shows the secondary communications device 64. The secondary communications device 64 are in communication with the primary communications device 46 in a direct connection configuration, a daisy-chain configuration or a combination thereof. Figure 5 shows an exemplary configuration of the primary and secondary communications devices 46,64. All of the communications devices 46,64 may use the same IP address while each camera 30 may use its own subnetwork, which simplifies and reduces the cost of the internet connectivity. Each secondary communications device 64 is configured to transmit image data from the respective camera 30 to the primary communications device 46 which in turns sends the image data to the computing device 32 via the cloud computing network 42. Each secondary communications device 64 is similar in structure and operation to the primary communication devices 46, except that each secondary communications device 64 does not require a communications controller 48. In other embodiments, the weight determination system may include multiple primary communications devices 46, wherein one primary communications device 46 is in communication with some of the secondary communications devices 64 while the other primary communications device 46 is in communication with the other secondary communications devices 64. In alternative embodiments, the computing device 32 may be located inside the shed or in the vicinity of the shed. In such embodiments, the computing device 32 may be configured to be in wired or wireless communication with the camera 30. In further alternative embodiments, the computing device 32 may be configured to receive image data from the camera 30 by way of a portable data storage medium. The computing device 32 as an image processor is programmed to process pixel information in the obtained top-down image of the chicken(s) to identify the chicken(s), perform a contact-free weight determination of the chicken(s), and generate a weight profile for the chicken(s) in accordance with a weight estimation algorithm. Although pixel count in the obtained top-down image can be used to estimate a size of a chicken which in turn is used to estimate a weight of a chicken, pixel count itself is insufficient to provide accurate weight data. This is because chickens of different ages, sex, breeds and health conditions can have different weights despite having the same size. Feather colour, patterns, feather coverage and feather texture visually differ between ages (e.g. chick, young chicken, adult chicken). Chicks and many chicken breeds have a uniform colour (with shadow areas), while some adult chickens with patterns can be taller but lighter than some other breeds with uniform colours (e.g. Ross 308). Chicks are usually yellow whereas many adult chickens have a red crest. Male and female adult chickens can have differently shaped and / or sized crests. Some breeds also have distinctive colour features, such as greyed and peppered white or whole brown (e.g. Hubbard), which are indicative of different weights for the samesized chicken. An exemplary step-by-step chicken weight determination process by the weight determination system of the invention is described as follows: The downwards-facing arrangement of the camera 30 enables a right-angle triangle to be formed between the camera 30, a centre point of the chicken underneath the camera 30, and a point on a border of the chicken. The formation of the right-angle triangle during image capture enables the use of trigonometry to standardise the geometry of images from the point of view of the camera 30. More specifically, the Thales theorem is used to normalise the geometry of images from the point of view of the camera 30, so that all images have the same geometrical aspects. In use, the camera 30 takes images of the chickens roaming freely in a live environment inside the shed, without having to restrict or force the chickens to a designated area. This can be further enhanced by using multiple cameras instead of a single camera, depending on the size of the shed. The camera 30 is configured to take the images continuously or at specific intervals, irrespective of whether a chicken is in the image or not. Alternatively the camera 30 may be configured to take images whenever a chicken is detected within an imaging region of the camera 30. It is not required to provide controlled environment settings, such as specific background colour or specific lighting, when taking the images. The camera 30 records an image, which may or may not include a chicken. The recorded image is then sent to the computing device 32 for processing by a weight estimation algorithm. The weight estimation algorithm then tries to find a chicken in the image. The algorithm's ability to find a chicken in the image is based on its training using data sets. The algorithm may be, but is not limited to, YOLO, Mask-RCNN and DeepLab. If no chicken is found, then the weight determination process is stopped, which means that the recorded image is not processed further. If one or more chickens are found in the image, the recorded image undergoes preprocessing to normalise the recorded image before being provided to an ML inference module of the weight estimate algorithm. The pre-processing may include one or more of the following: • Normalising the image pixels (in RGB, the value of each pixel is divided by 255 to generate pixel information within the 0 to 1 interval). • Resizing the image to a standard size (e.g. 300 pixels by 300 pixels). • Cropping the image to a fixed aspect ratio. • Zooming in or out of an image to normalise the distance from the shed's floor. • Perform pixel segmentation by reducing pixel values of the background to zero (i.e. black) so as to mask the background. Masking the background effectively removes the background while keeping the chicken(s) visible in the image. This allows the ML inference module to ignore the background and thereby focus on the chicken(s). Doing so reduces the amount of data processed by the weight estimation algorithm. Figures 6 to 11 compare original images and pre-processed images of chickens. Figure 6 shows an original image of a chicken weighing 790g while Figure 7 shows the same image after it has been pre-processed to resize the image and mask the background. Figure 8 shows an original image of another chicken weighing 850g while Figure 9 shows the same image after it has been pre-processed to resize the image and mask the background. Figure 10 shows an original image of a chick weighing 65g while Figure 11 shows the same image after it has been pre-processed to resize the image and mask the background. After pre-processing, the recorded image is provided to the ML inference module. It will be appreciated that, in other embodiments, the recorded image is not required to undergo pre-processing before being provided to the ML inference module. The ML inference module then processes the pixel information in the image to determine a weight of the chicken(s). The processed pixel information includes: pixel absolute values; geometrical relationships between pixels; and pixel count. Using pixel information in this way yields additional data, such as colour, brightness, feather patterns, feather coverage and feather texture, beyond only pixel count in order to take into account different sexes, breeds (e.g. Ross 308, Hubbard), ages, health conditions and other factors when determining the weight of the chicken(s). This in turn permits more accurate determination of the weight of the chicken(s). The weight determination algorithm can provide individual weight data for each chicken found in the image, as well as flock weight data such as average weight, top 10% values, bottom 10% values and coefficient of variation (CV%) as a representative value of flock uniformity. CV% is the distribution of weight within the flock and is calculated as a standard deviation divided by the mean weight of the flock. The weight data is then saved in the data storage medium. In addition, a weight profile, such as a histogram, of the chicken(s) can be constructed and saved in the data storage medium. Furthermore, the inventor observed that images of chicks have much fewer red pixels than images of adult chickens, and hence determined that red pixels could be used as a discriminator of age. Also, as a chicken's crest is typically red, the red pixel information can be used to not only identify the presence of a crest but also determine a size and / or shape of the crest so as to determine an age and / or sex of the chicken. To provide greater visibility of the red pixels, pixel segmentation may be performed to mask non-red pixels while keeping the red pixels. Figures 12 to 17 compare images before and after red pixel segmentation. Figure 12 shows a pre-processed image of a chicken with background masking while Figure 13 shows the same image after the non-red pixels are masked, leaving the red crest visible. Figure 14 shows a pre-processed image of a chicken without background masking while Figure 15 shows the same image after the non-red pixels are masked. In Figure 15, background artefacts remain present. Figure 16 shows a pre-processed image of a chicken with background masking while Figure 17 shows the same image after the non-red pixels are masked. Figures 14 and 16 are based on the same original recorded image. Optionally the weight estimation algorithm may convert the red pixels to white pixels to provide greater contrast, and therefore greater visibility, against the black background. The following table compares known weights of chickens shown in Figures 18 to 22 against estimated weights using the weight estimation algorithm. It can be seen that the estimated weights are very close to the known weights, which shows that the weight estimation algorithm is capable of producing accurate weight data from the images of the chickens. Image Known weight (grams) Estimated weights (grams) Figure 18 65 59 Figure 19 790 780 Figure 20 850 832 Figure 21 2045 1965 Figure 22 5320 5526 Figure 23 shows an original image of the shed floor recorded by the camera 30. It can be seen that there are multiple chickens in the image. Each chicken found in the recorded image is isolated to a single image normalised to a standard size (in this case 300 pixels by 248 pixels). Figures 24 to 26 show three isolated images of the chickens with background masking. Each image received from the pre-processing stage is processed by three artificial neural networks (ANN) in parallel as follows: a. The image is filtered with a convolutional neural network (e.g. ResNet-50) and then projected to a vector of size N floating point values (e.g. N = 615). This classifier searches for 2D patterns in images. The resulting classification guides the whole algorithm, e.g. to distinguish between breeds (uniform colours, striped animals, etc) and age (e.g. featureless chicks). b. A raw projection of the image to a vector of size R floating point values (e.g. R=463), and then a projection to another vector of size N floating point values. The raw projection encodes the original image to summarise the data. This notably encodes colour information global to the image, e.g. keeping track of yellow or white colours, indicative of chicks or adults. c. A matrix operation to perform red pixel segmentation, and then a projection to a vector of size N floating point values. Figures 27, 28 and 29 shows pre-processed images with red pixel segmentation corresponding to Figures 24, 25 and 26 respectively. In Figures 27, 28 and 29, the segmented red pixels were converted into white pixels for better visibility. A comparison of Figures 27 to 29 against Figures 24 to 26 shows that the segmented pixels correspond to the position of the crests on the chickens. The output of the three ANNs result in three vectors of size N floating point values, which are then added into V that is projected on a vector of size one. A sigmoid function is applied on the result to clip the value to a number in the 0-1 interval. The number is the output of the ML inference model and is used in combination with training data statistics to determine the weight of the chicken. Referring to Figures 24 to 26, the determined weights of the chickens are 3723g, 3765g and 4095g, with a weight range of 372g and an average weight of 3861g. ML training of the weight estimation algorithm can be exemplarily carried out as follows: 1. A dataset of pairs (image, weight) is provided. 2. The maximum weight is determined and stored for inference. 3. The dataset is normalised: a. The image's pixel values are divided by 255 so that each pixel is a number between 0 and 1. b. Each weight is divided by the determined maximum weight to obtain a number between 0 and 1. 4. The whole training subset is shuffled. 5. The dataset is split into training and validation subsets. 6. A training loop is executed: a. Apply a batch of B images to the weight estimation algorithm. b. Collect the B output normalised weights. c. Compute the differences with the B expected weights from the dataset. Differences are measured with a mean-square error standard algorithm. d. Apply the differences to the algorithm parameters using the standard back-propagation algorithm in stochastic gradient descent (SGD with momentum). e. Repeat until the dataset is exhausted to completed 1 epoch. 7. Evaluate and decide: a. Evaluate performance on the validation subset. b. If results are improving from any previous known run (e.g. lower error), go back to step 4 to perform the shuffling step before repeating the whole training loop. c. If results are not improving, go back to step 4 to perform the shuffling step before repeating the whole training loop. The return to step 4 is only performed for at most S times in a row. Beyond S times, stop training. 8. Final test: Evaluate the latest algorithm on test data for final scoring of the performance. The weight estimation algorithm may be led to learn from absolute pixel values in the images using several projection encodings on dense neural networks (e.g. so-called multi-layer perceptron). This is a set of typical direct linear regressions between categories like chick (yellow) and adult (not yellow), and different breeds (white, brown, grey). The weight estimation algorithm may integrate a range of backgrounds and lighting through data augmentation techniques during ML training. In terms of pixel colours, for example, this allows the weight estimation algorithm to be more robust to different feather colour aspects, which varies with lighting. Typically most of adult chickens are white but lighting makes them look blueish or orangish. Therefore, image temperature is integrated automatically in ML training procedures. Similarly, for backgrounds, specific ML training data covering target environments (such as wood chips on the floor, sand, protective sheets under chips / sand) are used. The listing or discussion of an apparently prior-published document or apparently prior-5 published information in this specification should not necessarily be taken as an acknowledgement that the document or information is part of the state of the art or is common general knowledge. Preferences and options for a given aspect, feature or parameter of the invention 10 should, unless the context indicates otherwise, be regarded as having been disclosed in combination with any and all preferences and options for all other aspects, features and parameters of the invention.

Claims

1. A weight determination system for determining a weight of an animal, the weight determination system comprising:an imaging device for obtaining an image of the animal; andan image processor programmed to process pixel information in the image to determine a weight of the animal, wherein the processed pixel information includes pixel values in the image.

2. A weight determination system according to Claim 1 wherein the processed pixel information includes pixel colour values in the image.

3. A weight determination system according to any one of the preceding claims wherein the processed pixel information includes pixel brightness values in the image.

4. A weight determination system according to any one of the preceding claims wherein the processed pixel information includes a pixel count in the image.

5. A weight determination system according to any one of the preceding claims wherein the processed pixel information includes a geometrical relationship between pixels in the image.

6. A weight determination system according to any one of the preceding claims wherein the processed pixel information includes a pixel pattern in the image.

7. A weight determination system according to any one of the preceding claims wherein the image processor is programmed to process red pixel information in the image to determine the weight of the animal.

8. A weight determination system according to any one of the preceding claims wherein the image processor is programmed to perform pixel segmentation on red pixels in the image prior to processing the pixel information in the image to determine the weight of the animal.

9. A weight determination system according to Claim 8 wherein the pixel segmentation includes reducing the pixel value of non-red pixels in the image.

10. A weight determination system according to any one of the preceding claims wherein the image processor is programmed to process pixel information in the image to identify a sex of the animal in the image so as to determine the weight of the animal.

11. A weight determination system according to any one of the preceding claims wherein the image processor is programmed to process pixel information in the image to identify a breed of the animal so as to determine the weight of the animal.

12. A weight determination system according to any one of the preceding claims wherein the image processor is programmed to process pixel information in the image to identify an age of the animal so as to determine the weight of the animal.

13. A weight determination system according to any one of the preceding claims wherein the animal is a bird.

14. A weight determination system according to Claim 13 wherein the image processor is programmed to process pixel information in the image to identify a feather, fur or hair pattern of the bird so as to determine the weight of the animal.

15. A weight determination system according to Claim 13 or Claim 14 wherein the image processor is programmed to process pixel information in the image to identify a feather, fur or hair coverage of the bird so as to determine the weight of the animal.

16. A weight determination system according to any one of Claims 13 to 15 wherein the image processor is programmed to process pixel information in the image to identify a feather, fur or hair texture of the bird so as to determine the weight of the animal.

17. A weight determination system according to any one of Claims 13 to 16 wherein the image processor is programmed to process pixel information in the image to identify a crest of the bird so as to determine the weight of the animal.

18. A weight determination system according to Claim 17 wherein the image processor is programmed to process pixel information in the image to identify a shape and / or size of a crest of the bird so as to determine the weight of the animal.

19. A weight determination system according to any one of the preceding claims wherein the image processor is programmed to perform pre-processing on the image prior to processing the pixel information in the image.

20. A weight determination system according to Claim 19 wherein the preprocessing includes: resizing the image; cropping the image to a fixed aspect ratio; and / or zooming in or out of an image.

21. A weight determination system according to any one of the preceding claims wherein the imaging device is arranged to, in use, obtain a top-down image of the animal.

22. A computer-implemented method of determining a weight of an animal, the method comprising the steps of:obtaining an image of the animal; andprocessing pixel information in the image to determine the weight of the animal, wherein the processed pixel information includes pixel values in the image.

23. A computer-implemented method of determining a weight of an animal, the method comprising the steps of:collecting a set of data, wherein the collected set of data includes pixel information in one or more images, wherein the pixel information includes pixel values in the or each image;creating a training set including the collected set of data;training a machine learning algorithm or model using the training set; and determining the weight of the animal based on an output of the machine learning algorithm or model.

24. A computer-implemented method according to Claim 23 wherein the machine learning algorithm or model includes an artificial neural network or a convolutional neural network.

25. A computer program comprising computer code configured to perform the computer-implemented method of any one of Claims 22 to 24.

Images