Abstract: Addressing the challenges posed by substantial throughput of raw cotton and prolonged impurity inspection duration in cotton mills, a solution was introduced wherein the algorithmic enhancement of the YOLOv7 model, incorporating lightweight modifications, was employed for impurity detection in raw cotton. Initially, redundant convolutional layers within YOLOv7 model were pruned, thereby optimizing detection speed. Following this, integration of Faster-Net convolutional layer within primary network mitigates algorithmic computational load, diminishes redundancy in feature maps, and consequently expedites detection process, enabling real-time capabilities. Ultimately, application of CSP-RepFPN (Cross Stage Partial Rep Feature Pyramid Networks) within secondary network facilitates reconstruction of feature pyramid, augmenting flow of feature information, minimizing feature loss, and elevating detection precision. Empirical results underscore enhanced YOLOv7 model's achievement of an aggregate detection accuracy of 96%, coupled with a 37.5% reduction in detection time. Accuracy of 82.5% on publicly DWC (Drinking Waste Classification) dataset with a detection time of only 29.8ms This refined YOLOv7 model presents a streamlined approach for impurity detection, yielding substantial time savings, and thus presents a viable avenue for discernment and classification of impurities in raw cotton.