AI Defect Detection

The AI defect detection for corn moldiness is a technology that utilizes human deep learning and computer vision to automatically, quickly and accurately identify the moldy areas on corn grains. Essentially, it is equipping the computer system with "eyes and brain" so that it can, like an experienced quality inspector, even more efficiently detect moldy defects in corn, and is widely applied in grain processing, storage and quality inspection fields. Through the real-time analysis of the trained deep learning model on images. The model has learned hundreds of thousands of normal corn and various moldy corn images, and can accurately identify the moldy colors (such as green, black, white mold spots), textures and other subtle features. Based on the analysis results of the AI, the system determines whether each corn grain is moldy, the type and severity of the moldiness. Once the system receives a signal, it immediately drives the actuator (usually a high-pressure valve) to precisely blow out the moldy corn and achieve separation of good and bad.

1) The comprehensive detection rate of various types of moldy grains is consistently above 99.8%, far exceeding the manual detection rate (approximately 70-80%). Through multispectral and deep learning technologies, interference is effectively eliminated, and the false detection rate is controlled below 0.5%, significantly reducing the waste of good grains.

2) Although there was initial investment in equipment, in the long run, it has significantly reduced labor costs, economic losses due to quality claims, and potential brand reputation risks.

3) It fundamentally eliminates the risk of moldy corn flowing into the downstream feed or food processing stages, significantly reduces the possibility of toxin contamination such as aflatoxin, and strengthens the food safety defense line. The moldy rate indicators of the output corn products can be quantified, improving product grades and market competitiveness.

4) The spatio-temporal distribution data of moldy rates generated by the system can help managers identify loopholes in the storage process, thereby achieving precise predictive maintenance and scientific grain storage, reducing the occurrence of moldy conditions from the source.

1) The optical properties of moldy corn under specific wavelengths differ significantly from those of normal corn. The system uses hyperspectral imaging to capture a rich spectrum of information beyond the visible light (RGB). It effectively distinguishes between moldy areas and normal color variations, as well as soil and germ layers, significantly improving the accuracy of detection. It particularly has a unique advantage in detecting early and subtle moldy conditions.

2) The model learns from thousands of labeled (normal, mild moldy, severe moldy) corn images to automatically extract deep features of moldiness, adapting to the appearance differences of corn from different varieties and origins. For the characteristic of early mold being only small dot-like, the feature pyramid network of the model is optimized to enhance sensitivity to small targets.

3) By rotating, scaling, changing brightness, contrast, and simulating different mold states, the training dataset is significantly expanded, improving the robustness and generalization ability of the model.

4) A uniform fabric device is designed to ensure that corn grains are uniformly and single-layerly distributed on the conveyor belt.

5) The detection results (moldiness rate, position images, timestamp) are uploaded in real time to the cloud platform. A quality report is generated, tracing the source of mold (such as a specific warehouse, batch), providing data support for warehouse management, and achieving early warning functions.

During the storage and transportation of corn, improper control of humidity and temperature can easily cause mold growth, which not only leads to economic losses but also poses a serious threat to the health of humans and livestock due to the toxins such as aflatoxin produced by the mold. Relying on the visual inspection by quality inspectors is inefficient, prone to fatigue, and unable to detect slight mold growth or internal mold. The standards are difficult to unify. To detect toxins, samples need to be sent to the laboratory, which takes several days and cannot meet the requirements for rapid circulation in grain processing and trade, nor can it achieve early warning. The mold forms are diverse (point-like, sheet-like, with varying colors), and it is difficult to distinguish them from the normal color spots and germ parts of the corn itself, resulting in high false detection rates and missed detections in traditional algorithms. To ensure food safety and improve quality control efficiency, we have developed this AI online mold detection system for corn processing enterprises.