Training Algorithms

Object detection is used to identify and locate specific visual elements such as defects or missing components within an image. 

By training on datasets that include both defect-free and defective parts, the system learns to recognise and outline each object with a bounding box label. 

Once trained, the model highlights detected defects and provides actionable outputs such as counts or pass/fail test results. 

Manufacturers can apply this module to automatically detect flaws or missing parts along the production line. 

The semantic segmentation module enables pixel-level classification of images, allowing Acumen AI to precisely identify and differentiate every region within a scene. 

Unlike traditional object detection, which outlines individual items with bounding boxes, semantic segmentation assigns a class label to every pixel, producing a detailed and comprehensive view of the entire image. 

This capability is ideal for applications that demand high-precision surface analysis, area-based measurements, or material differentiation such as coating inspection, surface wear detection, or contamination mapping. 

By delivering accurate pixel-by-pixel classification, the semantic segmentation module enables manufacturers to enhance inspection accuracy, improve process control, and accelerate decision-making through precise, visual insights. 

With semantic segmentation, Acumen AI delivers deep visual understanding of complex manufacturing environments driving smarter quality control and greater production efficiency. 

Instance segmentation enables precise pixel-level analysis by detecting and outlining irregularly shaped objects within an image. 

Unlike object detection, which uses bounding boxes, this approach employs polygon outlines to define object boundaries. 

The model differentiates between instances of the same class and isolates them from the background. 

The output includes a detailed mask showing each object’s shape allowing for accurate measurement, counting, and pass/fail evaluation. 

This technique is commonly used in the electronics industry to verify the presence or absence of conformal coating on PCBs. 

The classification module empowers Acumen AI to recognise, categorise, and sort parts or products based on their visual characteristics without the need to locate specific defect areas. 

By training on labelled image datasets, the AI learns to distinguish between predefined categories, enabling reliable identification of part types, variants, or surface conditions. 

This module is ideal for applications where quick and consistent sorting, verification, or quality assessment is required. 

Through deep learning-based feature extraction, it delivers rapid, automated classification results that eliminate manual judgment and improve process consistency. 

With the classification module, manufacturers can streamline inspection workflows, ensure product consistency, and enhance quality control through fast, accurate, and automated categorization. 

The result is a smarter, more efficient inspection process that brings speed, reliability, and intelligence to every production line

Anomaly detection identifies unexpected patterns or defects that differ from normal samples, even if those specific anomalies weren’t part of the training data. 

The model learns what constitutes a “good” image and flags deviations as “bad” or “abnormal.” 

Each image is assigned an anomaly score, indicating the severity of the deviation. 

During testing, normal samples are marked as passes, while anomalous regions are highlighted on a heat map for easy visual inspection. 

This approach is ideal for detecting subtle variations in shape, colour, or texture that fall outside expected tolerances. 

The anomaly detection module pinpoints subtle, previously unseen defects by learning normal patterns, enabling early detection of irregularities before they impact production quality.