Artificial Intelligence & Machine Learning

AVS AI ML Applying Machine Learning Title (MT)

Display portlet menu

Applying Machine Learning

AVS AI ML Applying Maching Learning Tabs

Display portlet menu
Illustration of a camera lens

Analyzing and understanding the content of images or videos is an important part of machine learning today. Deep learning models, and in particular convolutional neural networks (CNNs), are most commonly applied to analyze real-time visual image or video data in order to recognize objects, to classify images, or even to predict next actions in videos. Traditional computer vision or video pre-processing techniques can be applied in combination with machine learning techniques to accomplish your image and video projects.
 

 

Supported machine learning tasks

  • Image classification to describe the content of an image by its label or predefined category. It can be used to build up searchable image or video databases.
  • Object localization to locate the labeled content in the image or video with a bounding box. Object localization helps to understand what the object is, and where is it is located in the image or in the video.
  • Object detection to identify multiple objects of different categories within a picture or video. It can be used to find out how many objects of a certain class are in the image or the video and where the objects are located in the image for applications like people or object counting, or to make a distinction between objects, for example in a production process.
  • Semantic Segmentation associates each pixel of an image with a class label instead of marking objects with bounding boxes. Semantic segmentation helps to define the exact shape of objects and identifies each pixel that belongs to a person, a car, or any other entity in the dataset. You can group the entities based on color, texture, or other criteria. Semantic segmentation requires more processing power then object localization and detection, but it provides significantly more detailed information in images and videos.
  • Instance segmentation not only make pixel-wise predictions for objects as entities, but also identifies each entity separately. Instance segmentation can detect specific objects in an image and creates a mask around the object of interest.
  • Human Pose Estimation to identify the pose of a person by detecting the body parts. Human Pose estimation is useful for human computer interaction, video surveillance, or sports video analytics.

Solutions highlights

 

Deep Learning in Automotive on S32V234 Embedded Processors

Image & Video Anaylsis

AI Cloud with Microsoft Cognitive Services using i.MX8

Speech & Audio Processing | Image & Video Analytics

Machine Learning inference on the Ultra96

Image & Video Analytics

Artificial Intelligence Accelerated Camera Development System

Image & Video Analytics

Illustration of a magnifying glass looking at a bar chart

Machine learning algorithms have the ability to learn from data and to make predictions based on that data. Predictive analytics extract information from data sets in order to determine patterns and predict future outcomes and trends. Avnet Silica deliver solutions both for cloud and edge applications for industrial and consumer markets.
 

Supported machine learning tasks

Condition monitoring monitors parameters of the condition in machinery in order to identify changes of those parameters, which can be an indication for a fault or failure of the machinery. It helps to determine the condition of in-service equipment in order to estimate when maintenance should be performed and results in a reduction in unplanned downtime costs because of failure.

Several condition monitoring techniques are supported:

  • Predictive maintenance to evaluate the condition of equipment by performing continuous condition monitoring. The goal is to perform maintenance before the equipment fails or loses performance. Predictive maintenance benefits from preventive maintenance because it relies on the actual condition of equipment based on measured sensor parameters from multiple sensors.
  • Anomaly detection to identify rare items, events or observations, which raise suspicions by differing significantly from the majority of the data. Detecting anomalies over time helps to identify induced problems such as structural defects.
  • Prescriptive maintenance to automate complex decisions not only by detection and predicting upcoming issues or failures, like with predictive maintenance, but also by making advanced suggestions to reduce downtime or even to improve production.

Solutions highlights

 

Wireless Industrial Node for Condition Monitoring

Predictive Analytics

 
 
 
Illustration of Sound Waves

Analysis of sound is useful to identify the contexts related to audio signals, and to have an understanding of the environmental sounds. Speech & audio processing helps to analyze large amounts of natural language data to support the interactions between computers and human "natural" languages. This includes speech recognition, natural language understanding, and natural language generation.
 

Supported maching learning tasks

  • Natural language processing (NLP) to generate human-computer interactions. NLP includes speech recognition, natural language understanding, and natural language generation.
  • Acoustic scene classification (ASC) to categorize audio signals into predetermined classes. ASC is similar to speech recognition, except that the target classes are different and more heterogeneous. ASC can identify the contexts (classes) related to the audio signals and can be used in different applications including security and surveillance for gunshot detection, or just to identify the environmental situation you are in (restaurant, metro station, city street, ...).
  • Keyword Spotting (KWS) to detect predefined keywords in an audio stream. KWS is a technique to provide a hands-free interface for on voice-activated devices with limited resources and is used similar to AI voice assistants to detect wake words and keywords to trigger an action.

 

Solutions highlights

 

Smart Sensor with Machine Learning Core

Sound Analysis | Activity Analysis

Alexa Voice Control

Speech & Audio Processing

Offline Voice Control

Speech & Audio Processing

AI Cloud with Microsoft Cognitive Services using i.MX8

Speech & Audio Processing | Image & Video Analytics

Illustration of person in motion

Activity analysis is used to have an understanding of activity in observed scenes and is applicable to analyze walking patterns, optimize sport activities, or detect criminal or violent activity.  
 

Supported machine learning tasks

 

  • Activity Recognition to identify and predict the specific movements or actions of a subject, based on sensor data inputs. It is typically a time series classification task and involves methods from signal processing to extract features from the raw data to recognize the activity. Activity recognition can be used to identify human activities, such as walking, running, biking, driving, referred to Human Activity Recognition (HAR), or to monitor activity of animals for livestock management.
  • Human Pose Estimation can be used for activity recognition but will in addition identify the pose of a person by detecting the body parts by analyzing images or videos. Human Pose estimation is useful for human computer interaction, video surveillance, or activity recognition including sports video analytics.

Solutions highlights

 

Smart Sensor with Machine Learning Core

Sound Analysis | Activity Analysis