/AOI_Lab

Visual Insepction (include Anomaly Detect & Text Detection & OCR & BarCode Read)

AOI Lab(影像辨識與檢測)

Bulit on 2023/07 by Fred Liu
update 2024/02/20 (anomalydetection)

版本:MATALB: 2023a ~ 最新版本
需要工具箱: Deeplearning , Image Processing, Computer Vision, Parallel Computing
需要支援包: Computer Vision Toolbox Model for Text Detection,
Computer Vision Toolbox OCR Language Data,
Computer Vision Toolbox Automated Visual Inspection Library

Thanks for Alex Taylor Support anomaly detection

1.TextDetection

針對影像中的文字檢測並且標記，利用detectTextCRAFT模型。

AOI_TextDetection.mlx

image

2.OCR

針對文字做辨識，在2023a後的版本中，OCR的裡面的模型改為是Tesseract 5.0，演算法核心Deep Learning base， 架構為CNN+LSTM，所以整體精準度都有提高，目前也有62種語言與數字顯示器的辨識模型，並且可以在

AOI_DeepOCR.mlx
AOI_TrainDeepOCR.mlx
AOI_QuantizeOCR.mlx

image

3.BarCodeRead

針對一維與二維條碼，進行檢測與辨識。

AOI_BardcodeRead.mlx

image

4.Anomaly Detection

影像異常偵測與缺陷辨識，2022b之後的版本更新了三種異常偵測的演算法，分別是:

FCDD_Train.mlx
FastFlow_Train.mlx
PatchCore_Train.mlx

Network Function Notes
FCDD fcddAnomalyDetector
trainFCDDAnomalyDetector		 ．Light-weight model
．Fully convolutional
．Supports tiled training / full size inference workflow
FastFlow fastFlowAnomalyDetector
trainFastFlowAnomalyDetector		 ．State-of-the-Art model
．Fully convolutional
．Supports tiled training / full size inference workflow
．Relatively Memory intensive
PatchCore patchCoreAnomalyDetector
trainPatchCoreAnomalyDetector		 ．State-of-the-Art model
．Feature similarity based(no gradient descent training involved)
．Few-shot training
．Fixed image size at train and test
．Relatively Memory intensive(Supports compression)

image

5.Vit (Vision Transformer)

利用ViT網路進行影像辨識(Classification)

ViT.mlx

Vision Transformer Model有三種Pretrain Model

  • base
  • small
  • tiny
    image