SingTown's AI vision technology automatically identifies nuts, enabling roasting machines to precisely control the temperature.

In intelligent roasting equipment, different materials (such as peanuts, sunflower seeds, chestnuts, etc.) require matching with distinct heating curves and stirring strategies. Manual configuration is not only time-consuming but also prone to errors. To achieve automation and precise control of the roasting process, the roasting robot manufacturer plans to adopt SingTown Technology's OpenMV intelligent camera to build a machine vision-based material recognition system.

The system captures images of the raw materials inside the hopper via a camera, automatically identifies the type of food currently placed, and then calls the corresponding frying programme. The identification targets are primarily nut-based foods with a particle size of 1–5 centimetres. The camera is installed above the equipment and can capture the contents of the hopper within an adjustable distance. The system must be capable of distinguishing between different varieties such as peanuts, sunflower seeds, and chestnuts, maintaining stable identification even with minor positional deviations or angle variations. The objective is to enable the robot to automatically recognise the type of material, achieve unmanned loading and intelligent frying matching, thereby enhancing the equipment's level of intelligence and user experience.

The manufacturer of stir-frying robots, based on SingTown Technology's OpenMV intelligent camera material recognition system, integrates AI image recognition, colour and texture analysis, and machine learning classification algorithms to automatically identify material types in real time.

The system captures images of the material bin using high-resolution cameras and employs colour space conversion (RGB-HSV) and morphological feature extraction algorithms to analyse and classify the colour, texture, and shape characteristics of raw materials such as peanuts, sunflower seeds, and chestnuts. By training sample models, the system can rapidly identify the type of material currently in the bin, output corresponding identification signals, and automatically match the robotic stir-frying programme.

During the detection process, the OpenMV embedded vision module can automatically adjust exposure and white balance according to different lighting conditions to maintain stable and accurate recognition. Even if the material position in the hopper deviates or piles up, the system can accurately identify it through region segmentation and multi-feature fusion algorithms (colour + texture + contour shape). The recognition results are sent to the main control microcontroller via serial port or bus to achieve automatic parameter switching.

Measured data indicates that the system achieves an identification accuracy of over 95% for three common roasted snack ingredients (peanuts, sunflower seeds, chestnuts) under standard lighting conditions, with a response time of less than 0.5 seconds.

SingTown OpenMV intelligent cameras feature low cost, simple deployment, and strong anti-interference capabilities. They are suitable for edge deployment in various equipment such as roasting machines, nut packaging machines, and food sorting systems, facilitating the intelligent upgrade of the food automation processing industry.