Csmart Digit Desktop

User Guide for the Csmart Digit Desktop application

Overview

Csmart Digit Desktop is a desktop application for real-time coffee bean quality analysis using AI and computer vision. It connects to the Csmart Digit machine hardware to capture seed images, classify them using ONNX deep learning models, and produce detailed quality reports. Data is stored in HDF5 files containing images, extracted features, and classification results.

How It Fits Into the Csmart Ecosystem

Digit Desktop is the production analysis tool in the Csmart pipeline:

1. Csmart Digit Desktop captures images of coffee samples using the Digit machine and stores them in HDF5 format.

2. Csmart Studio Desktop uses those images to train and evaluate AI classification models.

3. The trained models (ONNX files) are deployed back to Csmart Digit Desktop for real-time inference.

Csmart Digit Desktop          Csmart Studio Desktop          Csmart Digit Desktop
(Image Acquisition)     -->   (Model Training & Export)  -->  (Production Inference)
       |                                                             |
  (Record Seeds)                                              (Classify Seeds)

Application Workflow

Digit Desktop is organized around a hub-and-spoke workflow. The Home screen provides quick access to all major functions:

Step	Screen	Purpose
1	Home	Create or open analysis files, access all features
2	New Analysis	Create a new HDF5 analysis file
3	Record Analysis	Capture seed images from the Digit machine camera
4	AI Classification	Run ONNX models on recorded images to classify seeds
5	Dashboard	View classification results, statistics, and charts
6	Image Mosaic	Inspect individual seed images in a grid layout
7	Lot Info	Enter metadata about the coffee sample
8	Classification Report	Export PDF reports with charts and statistics
9	Export Data	Export raw numerical data as CSV or Excel
10	Export Images	Export classified seed images organized by class
11	Blend & Compare	Compare multiple lots, simulate blends, track quality
12	Edit Model	Customize AI model parameters and classification rules
13	Evaluate Models	Compare AI model performance across analysis files

Additional screens manage Acquisition Settings, General Settings, Upload Analysis, Backup Settings, Runtime Environment, and About.

Key Concepts

Analysis Files

All analysis data is stored in HDF5 files (.hdf5). Each file contains the complete record of a coffee sample analysis: captured images, extracted features (color, shape, texture), classification results, and metadata. HDF5 is a high-performance binary format that efficiently stores large image datasets alongside numerical data.

AI Models

Digit Desktop uses ONNX models for seed classification. These models are trained in Csmart Studio Desktop and exported as .onnx files with an accompanying JSON metadata file. The JSON file defines class names, subsets, feature rules, density maps, and method equivalences.

Classification Modes

Seeds can be classified in two modes:

Mode	Description
Binary	Seeds are classified as OK or Defective (NOK)
Multiclass	Seeds are classified into specific defect categories organized by subsets: primary, secondary, foreign matter, and disregarded

Screen Sizes

Screen size refers to the sieve size used in coffee grading. It represents the physical size of the mesh holes through which beans are sorted. Digit Desktop tracks screen size for each seed (typically ranging from 10 to 19) and supports Moka (Peaberry) intermediates.

Moka (Peaberry) Beans

Peaberry (Moka) beans have a distinct round shape and different density characteristics. Digit Desktop supports separate recording and analysis of Moka beans, with dedicated density parameters for accurate weight estimation.

System Requirements

Minimum Hardware Requirements:

• Processor: Intel Core i5 (10th generation or newer) or AMD equivalent

• RAM: 8 GB DDR4

• Storage: 256 GB SSD

• GPU: NVIDIA GPU with CUDA support (recommended for faster inference)

• OS: Windows 10 or later (64-bit)

• Csmart Digit machine with Basler camera and ESP32 controller

Recommended Hardware:

• Processor: Intel Core i7 (12th generation or newer)

• RAM: 16 GB DDR4 or DDR5

• Storage: 512 GB SSD or larger (analysis files can grow to several hundred megabytes)

• GPU: NVIDIA RTX series (for GPU-accelerated inference)

• OS: Windows 11 (64-bit)

Prerequisites

Before using Digit Desktop, ensure you have:

1. The Csmart Digit machine connected via USB (camera) and serial (ESP32 controller).

2. An ONNX AI model trained and exported from Csmart Studio Desktop.

3. The Python runtime installed (from the Runtime Environment screen on first launch).

4. NVIDIA GPU drivers installed if using GPU-accelerated inference.

Getting StartedRecord AnalysisAI ClassificationDashboard