Introduction
The pig has played a major role in the food productivity for human consumption (Moeller and Crespo, 2009). Globally, pig meat is the second most preferred meat for consumption after poultry due to its low relative prices (Chatellier, 2021). China is leading the world in pork production in 2021, with more than 40 million metric tons, followed by the European Union and the United States (USDA, 2020). Also, pork consumption is projected to increase about 12% from 2017 to 2027 (VanderWall and Deen, 2018). Due to the fast expansion of the economy, demand for pork has expanded as well. The adoption of industrial farming practices for large-scale swine production is increasing (Huynh et al., 2006).
The construction of modern farms has been critical in meeting the growing demand for pigs. The growth of large-scale farming has presented new issues for breeders, since an increase in the number of pigs in pig farm results in a shorter monitoring period for each pig's living conditions (Yang et al., 2018). Among all of these challenges, it is vital to identify health problems and living situations that impact the most vulnerable pigs in order to preserve optimum health and animal welfare on commercial pig farms (Jorquera-Chavez et al., 2021). Major pig diseases are breathing or respiratory diseases and digestive diseases in pig farm. When a pig is infected with any of the disease pathogens, secondary disease complications might occur, resulting in health losses, productivity losses, and net profit at any stage of the pig's development. Without significant illnesses, better feed conversion efficiency may be used to promote a high quality of life while also increasing weight growth.
As the number of pigs in the farm rises, so does the density of pigs, and the influence of the pig farm environment on pig production becomes more and more obvious, especially for the purposes of monitoring. Various environmental characteristics in a pig farm, such as temperature, humidity, light, concentration of poisonous gases, and other elements, all have an influence on the pig's growth and development, reproduction, and final output performance (Han et al., 2017). Higher density in the pig farm also affects pig’s behaviour. It has been reported in various species of domestic pigs that their young have been injured or killed as a result of aggressive behavior, such as roughness, biting, assaulting, crushing, and death of the offspring (Houpt, 2018). Piglet crushing is another major concern in pig production. Highly dense pig condition caused the issue more vital. The mother pig squeezing the piglets, starving the piglets, poor viability, scours, and savaging might cause death loss. According to USDA reference statistics, 6 million new-born piglets are accidentally crushed beneath their mother sow every year (Andersen et al., 2011). Detecting and monitoring diseased pigs, as well as their odd behaviours throughout the infected period, and aggressive behavior in pigs, are becoming more challenging. As a result, automated identification of pig behavior is becoming more crucial, and it has emerged as a significant concern in the pig breeding industry (Han et al., 2017).
Image and video based sensing techniques can be used to collect and evaluate data in a manner that does not interfere with the pigs normal routine activities (Larsen et al., 2021; Garcia et al., 2020). In farms, image and video equipments enable in-depth monitoring of each individual pig. When compared to other technologies, image and video-based precision farming offers a number of significant advantages, such as real time monitoring, early alert system, etc. Furthermore, affordable sensors may be used to monitor without requiring direct touch, and a single sensor can be used to monitor a large number of pigs at the same time. Different implementation of image and video processing approaches like segmentation, separation of pig adhesion, extraction of pig, etc. are the feature parameters for automatic detection of diseases and visual symptoms in real-time to ease pig farm management (Chung et al., 2013). The main goal of image and video based pig diseases identification is to identify pigs. Color, texture, and shape are image properties of the object. Pig segmentation can provide an entire object outline. It is used to assess pig body size, weight, and posture. Static image segmentation uses feature classification to segment a single frame. Dynamic image segmentation uses the difference between frames to recognize moving objects.
Recently different methods were used to identify sick pigs and their symptoms based on the image and video processing. Object detection, Support vector machines (SVMs), and CNN-based methods have been very common methodologies used in recent years. Therefore, the goal of this review article was to provide an overview of recent studies dealing with the visual symptoms detection techniques in the pig farms for pig diseases and crush monitoring, including the aspects of detection methods for optimizing the technical parameters for detection.
Visual symptoms of pig diseases
Pig diseases visual symptoms
Pig respiratory diseases can be grouped into two main groups depending on the magnitude and duration: those that affect a significant number of pigs quickly and are potentially hazardous, and those that remain in a large number of pigs for an extended period of time (Fachinger et al., 2008). Sneezing, snuffling, and nasal discharge in piglets, as well as tear staining, twisting/shortening of the snout, and poor development, are the most visible signs of pig respiratory illness. Some disease were characterized by high fever, lack of appetite, coughing, pneumonia, convulsions, and blindness. Among the additional symptoms include nasal bleeding, slight snout deformities, shaking, twitching, irregular eye movements, convulsions, respiratory discomfort, conjunctivitis, coughing, trembling, head pressure, spasms, limb weakness, and death. Visual symptoms of different respiratory disease is shown in Fig. 1.
Digestive diseases develop when one or more entero-pathogens invade the bowel and cause significant anatomic or biochemical lesions that result in the passage of a watery stool. Enteric illness is a multifaceted issue. Pig digestive sickness caused by stomach ulcers manifests as in the form of diarrhoea, vomiting, abdominal discomfort, and/or dung with dark brown digested blood that resembles coffee powder. Diarrhoea is the most dangerous of all the illnesses that might occur. Growing pigs are more often impacted by this illness than mother pigs, who are less frequently affected. The visual symptoms of different digestive disease is shown in Fig. 2. Pigs are likewise arching their backs in discomfort, grunting, groaning, and sometimes screaming when they are experiencing intestinal problems. Pigs suffering from diarrhoea may acquire a high temperature of 42˚C, swelling of the neck, depression, vomiting, and an unwillingness to feed. It also causes difficulties to breath, and afflicted pigs may die within 24 hours of being exposed to it. In the case of recovered pigs, the dark skin may still be visible above the region of the swelling. Other signs and symptoms include paste-like or watery diarrhoea, push in the urine, weight loss, and dying suddenly. A jowl swelling, uneasy symptoms, a vacant abdomen, a cold and a languid disposition are all possible outcomes. Coma, convulsions, death, cough, and jaundice are other possibilities
Piglet crushing monitoring
The most common cause of piglet crushing and overlaying is the size disparity between the mother pig and the new-born piglet, which occurs most often when the mother pig is lying down to relaxation or to milk the piglets. Crushing may also occur as a result of a mother pig's sickness or behavioural disorder, which causes her to turn her back on the piglets. Because of the lack of distinct settings for new born piglets and large pigs on the farm, the crushing is evident. Piglets may also be crushed by themselves as a result of disease, hunger, hypoglycaemia, splay legs, joint illness, and other similar circumstances. Septicaemia is characterized by fever and sickness in the piglet. Crushing of piglets may also be caused by factors in the farm environment, such as insufficient separation of pigs and litter, lengthy straw bedding that prevents piglets from escaping, inappropriate rail or crate design, and low temperature and light levels.
The sound of a screaming piglet may indicate the presence of crushing. More commonly, dead piglets are discovered under the mother pig or with injuries that indicate that the piglets has been crushed. A lame or squealing piglet, or one with congenital tremor, splayed limb or restricted capacity to escape due to physical causes, may be diagnosed if the sound of screaming piglets can be heard. The most common symptoms of crushing are fractured bones, bruises, bleeding from the mouth or nose, death, and decreased viability in new-born piglets. Scratches from piglet teeth marks may be apparent on the udder, and there may be obvious indications of mastitis and agalactia on the surface of the udder. No movement, bites on ear and tails are also symptoms for the crush. It is possible to detect lameness in a sow by visual assessment, but it is often essential for the sow to get up before this can be verified. Behavioural issues in gilts are common, and they may manifest themselves in a variety of ways, including savaging or just bad management. Piglet failure may manifest itself as hunger, hypoglycaemia, splay leg, joint sickness, other likenesses, weakness at delivery, and cold, among other symptoms. The piglet crushing symptoms and different behaviour pattern is shown in Fig. 3.
Image and video based detection and monitoring systems for pig diseases and crushing symptoms
Camera sensors and softwares for visual symptoms detection Precision livestock farming is an intelligent technology that enables for more in-depth monitoring of each individual animal in farms. Current livestock production has a major and demanding issue in the area of disease monitoring and prevention. Early disease identification is an efficient method of averting large-scale disease outbreaks and their associated economic costs. Vision based system such as camera sensors, are widely accessible and can capture information more quickly than other types of sensor systems in pig farms, which is a significant advantage (Arulmozhi et al, 2021). Cameras are widespread instruments for research, and that have been available at a reasonable price around the globe. There are various kinds of cameras available on the market, including CCD cameras, infrared cameras, depth cameras, 3d camera and so on. Table 1 shows different types of camera used in animal farms for animal monitoring and surveillance. Each type of camera provides a unique set of information with picture and video characteristics (Arulmozhi et al, 2021). Early and automated identification of pig illnesses and behaviours is critical, and many kinds of cameras may be used to accomplish this. The use of a camera has become increasingly important in the pig breeding industry, as site tour inspections have proven ineffective in observing each pig's abnormal behavior while taking into account various factors such as radiation, floor type, growth stage, physical condition, nutritional status, health status, and other reasonable characteristics.
Table 1. Different types of RGB, depth and thermal cameras used in animal farms for animal monitoring and surveillance.
|
Any image-based analysis begins with image acquisition, which may be described as the process of obtaining numeric information using cameras (Fernandes, et al., 2020; Nasirahmadi, et al., 2017). The analysis of the pigs’ behaviours by image processing records an accurate behavior in a non-destructive way without disturbing their normal activities in real time (Larsen et al., 2021).
CCD cameras detect pixels of objects in red, green and blue (RGB) bands and convert them into grey, hue, saturation, intensity and other parameters by using different image processing algorithms (Nasirahmadi et al., 2017). CCD cameras measure the radiation of visible bands and thermal cameras detect the characteristic of near-infrared radiation with typically wavelengths of 0.75–1.4 µm and 8–15 µm, respectively (Rashman 2020).
Video processing is used to enhance the sound captured in video files and to alter the picture in accordance with the audio. The user may conduct editing activities by using a variety of filter algorithms. The output may be done either frame per frame or in bigger batches of frames (Revathi and Hemalatha, 2012). Different softwares and peripheral devices are used to load video files into the system, which allows the user to compile images and video, etc. by using the combination of pre filters, intra filters, and post filters signal processing (Chen et al., 2020).
The input image is initially pre-processed, which may include restoration, augmentation, or actual presentation of the real data, depending on the situation. Certain characteristics are retrieved from an image in order to segment it into its constituent parts (Verma and Sharma, 2013). The segmented images is then sent into a classifier or an image understanding system for classification and interpretation. Using image classification, distinct parts or segments of an image are classified into one of numerous objects, each of which has a label. In order to produce a description of an image, an image comprehension system must first detect the connection between distinct objects on a screen (Revathi and Hemalatha, 2012).
Image and video processing methods
Video denoising is the word used to describe the process of eliminating noise from a video stream. Spatial techniques, temporal methods, and spatio-temporal methods are the three types of methodologies. Each frame of video has noise reduction that has been applied. The visuals are altered as a result of the noise, and noise reduction is performed throughout the video processing process. The quality of the video processing approach is taken into consideration when determining the outcome of noise reduction. In the field of color image processing, a number of strategies for eliminating noise have been found.
Segmentation is the process of dividing a digital image or video into many segments in terms of sets of pixels, sometimes known as super pixels, and displaying them on a computer screen. In order to make things simpler and/or adjust the display of a picture into something understandable and easier to inspect, segmentation must be performed first. The process of segmenting an image or video is used to find the target object as well as its borders, such as lines, curves, and so on. Image performs the task of assigning a label to each pixel in an image and ensuring that pixels with the same label have certain visual qualities is pixels with the same label. When it comes to picture compression, image editing, or image database look-up, segmentation is utilized for object identification, occlusion boundary assessment inside action or stereo systems, and image database look-up.
Image analysis is the term used to describe the procedures that are used to extract information from an image. A grayscale picture is made up of the edges and contours of grey. Edge relates to a rapid shift in gray level and corresponds to information with a high frequency of repetition. The color shade corresponds to information with a low frequency. Edge detection is the process of separating high frequency information from low frequency information. The internal features in an image can be found using segmentation and texture. These characteristics are influenced by the reflectivity attribute. The process of segmenting an image involves isolating particular characteristics in the image. Rather of considering the remainder of the picture as a backdrop, if the image has a number of elements of interest, we may segment them one by one while treating the rest of the image as a background. The coarseness of an image's texture is a quantitative description of the texture. The coarseness index is proportional to the length of time that the local structure is repeated in space. A distinctive aspect of an image is referred to as an image feature. The most often utilized approaches for feature separation are those that operate in the spectral and spatial domains. The motions of an item are investigated via the examination of several photographs taken over a period of time ranging in length. Image analysis is distinct from other picture processing procedures such as restoration, enhancement, and coding, which all produce another image as a result of their operation. Image analysis is primarily concerned with the investigation of strategies for feature extraction, segmentation, and classification.
Several image-based and video-based studies have been conducted in recent years, using a variety of cameras and acquisition technologies to identify pig diseases, behaviours pattern and crushing symptoms. Using an image-based method, Weixing and Zhilei (2010) demonstrated a real-time pig breathing monitoring system. To identify the waist corner and scapular endpoint on one side of the ventral lines, RGB pictures were taken from the video clip and the Concave-Convex recognition technique was utilized to determine them. The enhanced chain coding technique is used to calculate the length of a line connecting two locations. In this case, the frequency of the curve might be used to describe the breath rate. With respect to determining respiratory rate, this paper's relative error is around 6.05% when compared to manual observation.
Huang et al. (2018) suggested a unique approach of identifying group-housed pigs based on Gabor and Local Binary Pattern (LBP) features. Texture features were recorded instead of the numerous color, shape, and texture information, which reduced the computational complexity and cost of the technique significantly. The Gabor wavelet was utilized to extract pig characteristics over a larger variety of scales, while the LBP was employed to capture minute appearance details in the pigs' look. Later, the feature dimension was reduced using Principle Component Analysis (PCA), and the features were concatenated to generate the feature vectors, which were then analyzed further. It was found that the combination of Gabor and LBP features provided greater discrimination, and the experimental results revealed that the suggested strategy may provide superior outcomes. The average recognition rate is 91.86% according to the data. Through the use of image processing, Kashiha et al. (2013) evaluated the possibility of an automated technique to detect marked pigs in a pen under experimental settings and for behavior-related research. Pigs were located using ellipse fitting techniques, which were implemented. As a result, using pattern recognition methods, individual pigs may be distinguished from one another based on their different paint patterns. Pigs could be recognized with a recognition rate of 88.7% when using visual labeling of videos were done by an expert. With the use of an algorithm for template matching and clustering algorithm, Li et al. (2019) demonstrated how to identify pigs in farm using a top view video surveillance system. Pigs could be detected with an accuracy of 86% using this approach. As a consequence, it is impossible to properly assess the approach since there are so many different ambient factors, image configurations and application of specific metrics need to take into consideration.
Zhang et al. (2018) suggested an effective on-line multiple pig detection and tracking approach that does not need pigs to be manually marked or physically identified and works in both daylight and infrared light situations. This technique combines a CNN detector with a correlation filter tracker through a unique hierarchical data association mechanism. It has been shown that the approach is capable of reliably detecting and tracking several pigs in difficult environments using data collected from a commercial farm. Promising results from the suggested technology also indicate a commercial viability for long-term monitoring of individual pigs in complicated environments. An efficient deep learning architecture was presented by Zhang et al. (2019) in order to directly identify pig behaviours in real time monitoring applications. The model increases the detection performance of the network by compressing the SSD network model, which means that the final two convolutional layers of the detection network are removed from the network. The algorithm was employed for real-time identification of three common sow behaviours: drinking, urinating, and mounting, all of which were observed. The algorithm correctly identified the drinking, urinating, and mounting behaviours with an average accuracy of 96.5%, 91.4%, and 92.3%, respectively. A real-time tracking and computation of animal movements is achieved using an algorithm developed by Fernández-Carrión et al. (2020). It employed a binary picture segmentation based on color threshold to identify the pigs, morphological operations and blob analysis for pig tracking, and Convolutional Neural Networks (CNNs) to count how many pigs were in each group. Video files were used to extract images. The results indicated a negative connection between mobility restriction and African swine flu induced fever. Additionally, diseased pigs were found much fewer movements than the healthy pigs. The findings indicate that an indoor motion monitoring system based on artificial vision might be utilized to raise concerns of fever. With the ultimate objective of establishing 24 hours continuous monitoring, Ju et al. (2018) focused on distinguishing touching-pigs in real-time utilizing low-contrast images from a Kinect camera to evaluate individual pigs using CNN-based YOLO algorithm. It was necessary to analyse and pick the best bounding box quality for each YOLO detector produced bounding box in the YOLO processing module. In real time, the approach was successful in accurately separating 91.96% of the touching pigs. However, the process is confined to two pigs that come into contact with each other.
For the identification of early warning symptoms of tail biting, Sonoda et al. (2013) proposed that automated video-based methods may be deployed. Recently, the use of 3D sensors to evaluate farm animal behavior has been considered. A Time-of-Flight 3D camera and machine vision algorithms were utilized by D'Eath et al. (2018) to automate the assessment of pig tail position in order to detect tail biting in pigs. The accuracy of 3D algorithm in recognizing low vs not low tails was tested and found to be 73.9%. Pigs were tracked down and oriented using proprietary algorithms and another method was used to find the tail of each pig that was present beneath the camera and standing up, and to measure the angle between the tail and the body on a scale ranging from 0 to 90 degrees for each pig. Lao et al. (2016) proposed and implemented an algorithm that continuously analyzed and assessed 3D depth images of sows laying, sitting, standing, kneeling, eating, drinking, shifting, and movement behaviours in farrowing crates using a low-cost 3D depth image sensor. The computational algorithm for the interpretation of depth images was applied, and the system showed high accuracy in classifying sow behaviours. The sow's movement activity increased as the piglets grew in size. There is still need to enhance the frequency of depth image collection in order to better record and quantify the sow's behavioural transitioning as well as better application in the pig farm. Martnez-Avilés et al. (2017) created an innovative smart system that uses a video surveillance camera to continuously monitor body temperature and movements in real time, allowing for the early identification of contagious diseases to be detected. It was found that pigs’ movement reduced as their body temperature rose, and this was analysed by automated analysis of video images, giving a less costly alternative to direct motion monitoring of the animals. To overcome data loss and numerous pigs’ observation, complicated circumstances and additional analysis with a larger dataset were required. To accurately separate individual pigs from a drinker and feeder zone, Guo et al. (2015) suggested an object extraction approach based on adaptive partitioning and multilevel thresholding segmentation, which they demonstrated on a pig farm. The top view method was utilized to analyse group-housed pigs in complicated situations, independent of the pig numbers, color, sluggish motions, or behavior of the pigs. The average detection rate was 92.5%, according to the findings of the study.
According to Yu et al. (2021), a computer vision system was developed by collecting RGB-D videos to record the top-view of pigs and depth images of growing pigs in order to forecast their body weight over time. To process the images, a Python-based image-processing algorithm based on OpenCV was created. A thresholding method was used to segment each pig inside the RGB frame, in particular. The prediction coefficients of determination range between 0.49 and 0.95, depending on the mixed models used to make the predictions. It was discovered that nonstationary imagery, the distance between the sensors and the pigs, pigs contacting each other, touching the fences, and sitting situations, constituted a significant challenge for the proposed approaches, and these challenges impacted the segmentation results. Two depth cameras were used in the development of a portable and automated pig body size assessment system by Wang et al. (2014). RANSAC, a random sample consensus algorithm, was utilized to analyse the background point cloud and obtain the foreground pig point cloud, which was then clustered using Euclidean distances. The average relative errors for body width, hip width, and body height measures were found to a range between 5.8% and 10.30%. As a result of the uneven ground surface, misclassifications occurred, and the point cloud separation procedure was hampered. A technique for identifying low-weight pigs using moving pigs' body sizes has been proposed by Sa et al. (2015) with an automated system. The approach did a binarization of each frame in order to conduct background removal from a video collected by a surveillance camera using a video acquisition system. It also recognized two pigs in the binary image using the Gaussian Mixture Model (GMM), to detect them. Calculated from the amount of pixels in each moving pig, the average size of the moving pigs were measured. The low-weighted pig threshold was set at 70% of the average, and the approach was able to identify low-weighted pigs as a result. Multiple pigs and higher dense pig condition could be challenging for the results accuracy. Different image processing methods for pig diseases detection and tracking and monitoring of pig in farms is shown in Fig. 4.
Fig. 4. Different image processing methods for pig diseases detection and tracking and monitoring of pig in farms, (a) single pig detection (modified from Huang et al. (2018)), (b) multiple pif detection (modified from Kashiha et al. (2013)) , and (c)group of pig tracking (modified from Li et al. (2019)).
Using infrared thermography (IRT), Lu et al. (2018) proposed an approach for automatically extracting ear base temperature from top view piglet thermal images. For the identification of piglet head parts, a Support Vector Machine (SVM) classifier was developed. Then, using the shape characteristic of the head portion contour as a guide, two ear base points were determined. Finally, the ear base temperatures were determined by extracting the two highest temperatures that occurred inside the two circles that were cantered on the ear base locations. According to the findings, 97% and 98% of the evaluated images had an inaccuracy of less than 0.4°C for the left and right ear bases, respectively. Due to the different positions of the head and ear in the several photos, the incorrect ear temperature was also seen. Tabuaciri et al. (2012) developed a successful approach for diagnosing hypothermic piglets, as well as a feasible alternative to assessing the real core body temperature of new-born piglets in the laboratory. In terms of rectal temperature and IR values, the mean temperatures of shivering and non-shivering piglets were substantially different throughout the three anatomical locations examined. It was discovered that the mean temperatures measured at the tips of the ears were rather low, and that measuring them was extremely difficult. All shivering piglets should be given immediate treatment; however, those with low infrared body temperature within 24 hours after farrowing should also be given further care. Caldara et al (2014) investigated the influence of the environment on the weight of piglets at birth using IRT imaging system. The authors estimated the heat loss of the piglets in the first few hours after delivery and revealed the importance of maintaining a comfortable temperature in the birth place. There is a significant amount of thermal exchange through conduction between new-born piglets and the farrowing crates’ floor (Fig. 5), and this may have an impact on piglets' body heat loss, reducing their performance. Through the use of IRT images, that piglets with lower birth weights have a greater drop in body temperature in the first hour of life than other piglets. To identify edema in pigs, Graciano et al. (2014) employed IRT image in the pig farm. The authors demonstrated the effectiveness of IRT image analysis in the diagnosis of arthritis in its early stages.
The time required for large-scale phenotyping of pig disease, behavior, and shape and size features has boosted demand for technology that automate these operations. While automated tracking and detection of pig has been effective in controlled laboratory settings, recording from large groups of animals in highly varied agricultural environments poses difficulties (Wurtz et al., 2019).
Challenges and perspectives of visual symptom process
The use of cameras for pig diseases detection, behavior monitoring, health condition and so on has already been applied to pig and other animals that are easy to manage in highly-controlled settings and large amount of animals (Wurtz et al., 2019). Image and video surveillance in pig farms provide huge advantages for the monitoring works in pig farms. A variety of obstacles are presented by commercial farm settings, including clusters and feeding habits, unmarked animals, different lighting and environmental background, and the likelihood that the animal gets filthy with debris or faces. There are numerous challenges in the automatic recording or image acquisition of pigs in the farm when using camera sensors and other optical sensors.
Individual pigs or group of pigs may be detected and monitored using image and video surveillance systems, but there are significant challenges (Zhang et al., 2019) such as:
Variability of illumination. For example, abrupt light changes regularly occur in pig farms, including varied illuminations throughout the day and night, which might cause the camera imaging model to alter. This results in inevitable shadowing under various lighting situations in image and video files.
Pigs with very similar features and varied levels of pig background. In image or video files, the size, shape, and color of the pig in a pan or under the same shed are almost identical to each other. Furthermore, the varied pig background in the farm has made image processing more complex in various ways.
Deformations and occlusions of objects. Bugs temporarily obstruct the lens, and group of pigs obstruct one another. Although individual pig identification and tracking are considered to be the most critical steps in the majority of image and video-based pig monitoring applications, creating a system that is capable of coping with these circumstances remains an unsolved challenge in this situation.
To increase our capacity to automatically characterize behavior, future research should build on current knowledge and verify technologies in commercial contexts, while publications should fully explain recording circumstances in sufficient detail to permit replication of findings (Wurtz et al., 2019).
Conclusion
Current research on image and video-based automated pig disease identification and piglet crushing symptoms recognition algorithms was summarised in this study. The identification of pig diseases and piglet crush monitoring has emerged as a research priority in the area of intelligent livestock production. The approaches that were employed for pig segmentation, pig detection, and behavior identification are discussed in further detail. Both standard image processing techniques and deep learning approaches are used to process images. Up to this point, some excellent work has been accomplished; nevertheless, there is still much more work to be done to enhance image acquisition techniques, cost, time and methods accuracy in the detection of pig illnesses and crushing symptoms. Artificial intelligence based learning has rapidly gained popularity and has progressed from basic models to sophisticated models as a result of technological advancements. The immersive adoption of the described technology will contribute significantly to pig production and will assist in overcoming some associated limitations through recent technology such as the use of cameras and sensors, an affordable price, dependability, ease of handling, and other advanced technology such as drones, robots, and other advanced technologies.
Acknowledgement
This work was supported by the Korea Institute of Planning and Evaluation for Technology in Food, Agriculture and Forestry (IPET) through Smart Farm Innovation Technology Development Program, funded by Ministry of Agriculture, Food and Rural Affairs (MAFRA) (Project No. 421044-04), Republic of Korea.
