The field of machine vision in manufacturing is rapidly evolving, with increasing demands for higher precision and efficiency. As expectations rise, manufacturers face significant challenges in maintaining speed and accuracy, especially when dealing with high-resolution images required for detecting small defects. Analyzing such large images in real time often leads to increased inference times, posing a critical challenge for vision inspection systems.
To address these complexities, Neurocle has launched version 4.1 of its deep learning vision inspection software, tailored to meet the nuanced needs of modern manufacturing environments. The new features promise to tackle various challenges, such as detecting defects in high-speed production lines, managing limited defect data, and optimizing performance on lightweight embedded devices.
Neurocle’s advanced capabilities span the entire vision inspection workflow, powered by its core technology: Auto Deep Learning. This cutting-edge algorithm automatically optimizes model architectures and hyperparameters, simplifying the training process and enabling users to develop high-accuracy inspection models without deep learning expertise. Neurocle’s model creation platform (model trainer) Neuro-T, ensures that even complex manufacturing scenarios are addressed with ease.
Enhanced High-Resolution Defect Detection with Patch Classification (PAC)
One notable feature in the 4.1 release is the introduction of the Patch Classification (PAC) model, designed to overcome the limitations of traditional classification models that struggle with high-resolution image data. Previously, resizing high-resolution images for analysis often led to a loss in detecting defects. This model segments high-resolution images into smaller patches, ensuring that even the tiniest defects are accurately detected without compromising on resolution.
Improved Data Augmentation with GAN Patch Mode
For situations where defect data is scarce, Neurocle has enhanced its Defect Generator (GAN) model capabilities. The new GAN Patch mode slices images into patches to generate more precise synthetic defect images. This advancement facilitates the adoption of deep learning inspection in manufacturing environments with limited initial defect data, reducing the burden of data collection and accelerating the implementation of deep learning-based inspection systems.
Efficient Labeling with Shape Converter and Auto-Labeling Tools
Neurocle’s Shape Converter and Auto-Labeling tools provide innovative solutions for meticulous labeling tasks. The Shape Converter transforms box labeling areas into detailed
polygon shapes with a single click, while the Auto-Labeling tool applies consistent labeling standards across all images after labeling just a few samples. These features significantly reduce labeling resource requirements and enhance operational efficiency.
Expanded Processor Support for Diverse Inspection Environments
With support for OPENVINO, DirectML, GPU, CPU, NPU, and embedded boards, Neurocle’s software is capable of integrating into various manufacturing inspection setups. This broad processor compatibility ensures that manufacturers can deploy deep learning models across a wide range of hardware environments, enhancing flexibility and application potential.
Neurocle envisions itself as the Adobe or PowerPoint of the vision inspection industry, offering a comprehensive tool with all necessary functionalities. The company continually anticipates diverse scenarios that manufacturers might encounter during the inspection process, striving to develop solutions that address these challenges effectively. The user-friendly and highly advanced features of the new version 4.1 will be showcased at Vision Stuttgart this October, setting a new benchmark in vision inspection technology.
By offering tailored solutions for specific manufacturing challenges, Neurocle’s 4.1 release redefines the capabilities of deep learning vision inspection, enabling manufacturers to achieve unprecedented levels of precision and efficiency.
