Test the model directly on your browser and use our Python SDK to build your application ready to run on your own device. Solve your classification, detection, segmentation, or person recognition problems executing the model in the edge on real time.
Count and track animals, detect weed seed, recognize people, detect defects, etc.
Easily annotate your images with DeepLab, and train a foundation model for your application. Select the task and run your own notebook to easily train a model, or request assessment for your application.
Consult your problem and get a solution ready to be tested on the field.
From Abraia Vision we develop custom solutions for several problems which are susceptible of being solved using state of the art deep learning models and image processing. Our solutions focus on models which can be run in the edge, on your own hardware, without the cloud.
AI vision systems allow information extraction with minimal external inferences (human adjustment of sensors, maintenance) at an affordable cost. So, by combining computer vision with the power of artificial intelligence we can develop smart solutions which can reduce production costs and increase productivity.
Compared to traditional computer vision, modern deep learning algorithms are much more robust and allow highly accurate real-time image recognition. Hence, deep learning methods can be used to perform video analytics with the video of common surveillance cameras or webcams.
The latest trends combine edge computing with on-device Machine Learning; a method also called Edge AI. Edge-AI brings computer vision to portable devices. Lightweight models can run on mobile hardware to analyze plant images and detect early signs of disease. This reduces dependency on cloud systems and enables offline decision-making. Moving AI processing from the cloud to edge devices makes it possible to run machine learning everywhere, combining the Internet of Things (IoT) and AI to create scalable computer vision applications.
Computer vision models for image classification, object detection, image segmentation, and pose estimation, provide the foundations for building custom computer vision solutions for multiple problems and sectors. For instance, computer vision in agriculture enables intelligent agriculture and food production for smart farming applications.
1Traditional disease identification relies on manual scouting, which can be time-consuming and prone to human error. Image recognition and classification enables rapid, large-scale disease detection, helping to prevent outbreaks before they spread.
Moreover, drones equipped with multispectral and hyperspectral cameras can scan entire fields, enabling identifying areas of plant stress days before symptoms become visible to the human eye. Detecting variations in plant health, soil moisture levels, and nutrient deficiencies across vast farmland. For instance, NDVI (Normalized Difference Vegetation Index) imaging helps farmers assess plant vigor by analyzing how crops reflect light.
Based on drone footage and geo-sensing information, crop mapping and planning enables field monitoring and management. This can be used for predicting overall crop yields by analyzing the number of mature fruits or vegetables present in a field, giving farmers a more accurate forecast of their harvest.
2Yield estimation is an essential preharvest practice, supporting decision-making for allocating essential logistics such as transportation means, labor force, supplies, and more. An overestimation leads to further costs that impact profitability; understimation entails potential crop waste and additional costs.
Manual yield estimation with the counting of products such as fruits or vegetables is very time-consuming and expensive. Modern deep learning methods provide good accuracy for automatic counting, even with occlusions caused by leaves or branches, illumination, and object size. So, computer vision approaches can be used for the automatic counting of fruits or flowers. An example of this is the automatic on-tree counting and yield estimation of kiwifruit.
3Computer vision systems can capture detailed images of produce as it moves along a conveyor belt. Then, AI-powered algorithms can analyze these images to assess critical quality factors like size, shape, color, texture, surface defects, and ripeness.
These agriculture computer vision systems can sort produce like potatoes or apples based on order requirements on large volumes, and used for quality control. With computer vision technology it can be achieved high accuracy in detecting defects and foreign materials, enhancing product quality, and reducing waste.
For example, in apple grading, computer vision systems evaluate the size of the apples, check for skin blemishes, and assess color consistency to classify them into various quality grades. These systems can be trained to detect subtle defects like bruising, rot, or pest damage that might be missed by the human eye. That way damaged or subpar items can be removed before they reach consumers.
4Animal monitoring systems (based on pose estimation models) provide continuous real-time monitoring and assist producers in management decisions. AI vision is able to provide objective measures of animal behaviors and phenotypes as opposed to subjective manual observation.
Smart vision systems can be used to monitor animals such as cattle, sheep, pigs, or domesticated species, including chickens, turkeys, ducks, geese, and others with cameras under field conditions. Neural networks are used to analyze video feeds in real time. The advantages of computer vision systems are rooted in their automatic, non-invasive, and low-cost animal monitoring capabilities.
Moreover, modern methods are capable of assessing the resources provided to the animals (space, lying substrate, drinker access) and measuring the animals themselves to detect lameness, indicators of injury or disease, and abnormal behaviors. Hence, computer vision provides quantifiable data about animal welfare that can be used to ensure compliance with on-farm animal welfare.
5Real-time surveillance and security monitoring for remote farms can be achieved using modern deep neural networks to perform accurate face recognition that is invariant to changes in illumination. This makes it possible to implement deep face recognition in multiple remote farms. Such monitoring and notification systems are of high importance to farms. The images detected with common surveillance systems can be processed by AI algorithms to perform intrusion detection and automatically identify anomalies.