A Solution For Anomaly Detection of Red Beans In A Product Processing Line

Anomaly detection (AD) plays a crucial role in ensuring the quality and safety of processed products. With the rapid growth of machine learning, several AD frameworks and datasets have been introduced in the past decades. However, the ability of those models to adapt to real-world settings and overcome practical hindrances remains an issue. In this paper, we introduce a solution for abnormal red beans detection on a conveyor belt of the product processing line. We aim to provide deeper insights into the features and challenges of this task and introduce an end-to-end method which, to our best knowledge, is the first model to be applied to our regarded settings. The experimental results of our model show its promising potential for real-world industrial AD, both in terms of accuracy and latency.