AI Vision Inspection for Food Manufacturing: Deep Learning Solves Overlapping Betel Nut Counting Challenges

Accurate product counting is essential in food manufacturing, directly affecting packaging quality and production efficiency. Traditional photoelectric sensors work well when products are separated, but they struggle whenever items overlap or stack together.

A food manufacturer in Wuhan faced this exact challenge on its betel nut production line. By replacing conventional sensor-based counting with an AI-powered machine vision inspection system, the company achieved highly accurate counting even when multiple betel nuts overlapped on the conveyor.

This case demonstrates how deep learning object detection can significantly improve counting accuracy and support smart factory automation.

Betel Nut Counting Dynamic Effect

The production line originally used a vibrating feeder combined with photoelectric sensors to count betel nuts before packaging.

However, during continuous production, two or more betel nuts frequently overlapped or touched each other while moving along the conveyor belt.

Because photoelectric sensors generate only a single signal when multiple products pass together, the system often counted several betel nuts as one, resulting in:

• Incorrect product quantities
• Missing or extra items inside packages
• Increased defect rates
• Reduced production quality
• Higher manual inspection costs

To eliminate these problems, the manufacturer implemented an AI vision inspection system based on deep learning object detection, replacing the traditional counting method.

Betel nuts continuously roll and rotate on the conveyor, creating significant variations in:

• Orientation
• Shape
• Surface wrinkles
• Partial occlusion
• Overlapping positions

These variations make traditional rule-based machine vision algorithms unreliable.

The conveyor belt frequently accelerates and decelerates during production.

Changing speeds introduce:

• Motion blur
• Position shifts
• Inconsistent object spacing

The vision system must accurately detect moving targets under dynamic conditions.

Food production lines require continuous real-time inspection.

The AI model must deliver:

• High detection accuracy
• Low inference latency
• Stable long-term operation

without slowing down the production line.

The inspection system uses a deep learning object detection model trained specifically for overlapping food products.

Unlike traditional sensors, AI identifies every individual betel nut independently, even when several products are touching each other.

The solution includes:

• Deep learning object detection for overlapping product recognition
• Image enhancement to reduce motion blur
• Optimized feature extraction for rotating objects
• Lightweight AI model for real-time inference
• Industrial hardware acceleration for continuous production

The system accurately separates overlapping products and performs precise counting without affecting production speed.

The AI vision system detects each individual betel nut independently, eliminating counting errors caused by overlapping products.

Optimized image processing minimizes the effects of conveyor acceleration, deceleration, and product movement, ensuring stable inspection performance.

A lightweight deep learning model reduces inference time while maintaining high detection accuracy, making it suitable for high-speed food manufacturing.

The same AI vision technology can be quickly adapted to various food counting and inspection applications, including:

• Instant noodle fork detection
• Snack counting
• Grain and seed inspection
• Candy counting
• Small packaged food inspection
• Loose food product counting

Deep learning machine vision is becoming an ideal replacement for traditional photoelectric and infrared sensors in many food processing applications.

Typical use cases include:

• Product counting
• Missing item detection
• Packaging verification
• Multi-object counting
• Conveyor inspection
• Quality inspection
• Automated sorting
• Smart manufacturing

Compared with conventional sensor-based solutions, AI vision systems provide higher accuracy when products overlap, rotate, or appear in irregular positions.

As food production lines continue to increase in speed and automation, traditional sensor-based counting systems are no longer sufficient for handling complex product arrangements.

By leveraging deep learning object detection, AI vision inspection systems can accurately identify, separate, and count overlapping products in real time while maintaining high production efficiency.

This project demonstrates that AI-powered machine vision is an effective solution for improving counting accuracy, reducing packaging defects, and enabling smarter, more reliable food manufacturing processes.

Yes. Deep learning object detection can recognize and count individual products even when they overlap or touch each other, significantly outperforming traditional photoelectric sensors.

Photoelectric sensors detect interruptions in a light beam. When multiple products pass together, they often generate only one signal, leading to undercounting or missing products.

AI vision inspection systems are widely used for counting betel nuts, instant noodle accessories, snacks, candies, grains, seeds, frozen foods, packaged food components, and many other products moving on conveyor belts.

Yes. Modern lightweight deep learning models combined with industrial AI hardware can perform real-time inspection without reducing production throughput.