K64C Vertical AI Coffee Color Sorter

Vertical AI Coffee Color Sorter for Compact High-Precision Lines

The K64C is a single-chute vertical AI color sorter designed for coffee processing where floor space is limited but detection detail still matters. Two intelligent 2 MP area-array cameras per chute read each bean by shape, size, texture and color, while a dedicated 16 TOPS NPU per chute runs the recognition model in real time. High-speed matrix solenoid valves fire the rejects at 1200 Hz, giving accurate ejection without disturbing the good-bean flow.

AI Vision & Deep-Learning Architecture

Imaging, AI inference and ejector control are integrated into one high-speed sorting architecture. Each chute pairs two smart cameras with two NPU computing boxes for low-latency decisions:

• High-resolution AI imaging — 2 MP area-array capture reads shape, size, texture and color.
• Zero-distortion imaging — correction technology preserves geometry, edge detail and defect location.
• Parallel NPU acceleration — 16 TOPS per box, up to 32 TOPS per chute.
• On-machine model training — label images and train the sorting model directly on the 15.6-inch touchscreen.
• Remote technical assistance — online debugging and model training from a computer or smartphone.
• Adjustable RGBW lighting — red, green, blue and white illumination with adjustable color and brightness.

Material moves through four steps: dual cameras capture the flow, the NPU models analyze and classify each piece, 1200 Hz valves eject rejects, and accepted material exits through the discharge conveyor below.

What the K64C Sorts

Beyond coffee, the same deep-learning platform handles other granular materials where color-only sorting is not enough:

• Coffee beans — color defects, shape differences, surface damage and foreign material.
• Rice & grains — discoloration, shape defects, impurities and inconsistent kernels.
• Pulses & seeds — color, size, texture and shape-based classification.
• Plastics & other granular materials — multi-feature visual sorting, subject to sample testing.

Why Processors Choose the K64C

• Multi-feature AI vision — detects defects by shape, size, texture and color, not color alone, so immature, broken and insect-damaged beans are caught alongside black and sour beans.
• 32 TOPS per chute — two 16 TOPS NPU boxes per chute keep inference fast even at full 1-2 t/h feed.
• 1200 Hz valve array — 8-nozzle blocks respond in ≤ 0.06 ms with a service life of 10 billion cycles.
• Fast model setup — a typical sorting model is trained from about 50 labeled images in roughly one hour.
• Remote-ready control — 15.6-inch capacitive touchscreen, built-in Wi-Fi and a Linux industrial controller support remote technical assistance.

Technical Specifications

Note: Capacity and sorting performance vary with material characteristics, defect ratio, feed condition and target quality. Final configuration is confirmed through sample testing — send us your material samples and sorting requirements.