Lightweight lotus seedpod quality classification algorithm based on improved YOLOv3-Tiny model

An accurate and efficient quality classification algorithm for lotus is an important part of realizing the automatic processing of lotus after harvest. Regarding the lack of studies on the quality classification of lotus after harvest, this study established a lotus quality classification principle based on the evaluation scheme for white lotus quality published by the Chinese Association of Traditional Chinese Medicine, and proposed an improved YOLOv3-Tiny (You Only Look Once version 3-Tiny) model for quality classification. First, a vertical camera was deployed based on the principles of image recognition to establish an image acquisition platform for lotus seedpod, enabling the collection of lotus seedpod images under three different lighting conditions and the creation of a comprehensive test dataset, the dataset was then augmented using three image augmentation techniques: rotation, colour switching, and affine transformation. The K-means clustering algorithm was subsequently employed to optimize the scales of the prior anchor boxes, aiming to enhance the regression accuracy. Ultimately, six distinct scales for the prior anchor boxes were generated, namely (14, 18), (19, 21), (20, 27), (36, 44), (42, 49), and (50, 56). The achieved accuracy rate is 85.48%. Subsequently, based on the original YOLOv3-Tiny backbone network, the SPP (Spatial Pyramid Pooling) module was added to improve the network’s ability to extract feature information. Finally, the parameter evolution module of YOLOv3-Tiny was used to evolve a set of appropriate hyperparameters for the model. To verify the effectiveness of data augmentation, redesigning the prior anchor box scales, optimizing the feature extraction backbone network and related hyperparameter evolution in improving model performance, ablation experiments were carried out. The ablation experiments showed that recognition precision is increased by 4.95% by using the data augmentation technique. Using the K-Means algorithm to reunite classes, the recognition precision increased by 0.76%, Based on the K-Means algorithm for class reunion, adding the Spatial Pyramid Pooling (SPP) module further improved the recognition precision by 1.12%, By adding the hyperparameter evolution module, the recognition precision reached 93.1%, and the overall robustness of the model improved. To verify the effectiveness of the improved YOLOv3-Tiny model on the quality classification of lotus seedpod, we tested its performance against the original YOLOv3-Tiny network model and Faster R-CNN. The results show that the improved YOLOv3-Tiny model achieved a Mean Average Precision (mAP) of 96.80%, a Precision (P) of 93.10%, and a Recall (R) of 94.60%. Compared to the original YOLOv3-Tiny model, the mAP is increased by 12.49%, the Precision (P) is increased by 14.77%, and the Recall (R) is increased by 11.59%. The number of frames transmitted per second reached 25 Hz, which is 1.24 times that of the Faster R-CNN network model. Ablation experiments showed that the Precision (P), Recall (R) and mAP of the improved YOLOv3-Tiny model increased by 4.95%, 4.52% and 5.61%, respectively. By redesigning the prior anchor box scale, the Precision (P), Recall (R) and mAP of the network model were 0.76%, 0.10% and 0.25% higher than those of the model with the original prior anchor scale. By introducing the SPP module, the Precision (P) increased by 1.12%, the Recall (R) increased by 0.10%, and the mAP increased by 0.23%. Through hyperparameter evolution, the Precision (P) increased by 7.94%, the Recall (R) increased by 6.87%, and mAP is increased by 6.40%. Ablation experiments verified that data augmentation, redesigning the prior anchor box scale, improving the YOLOv3-Tiny feature extraction backbone network, and related hyperparameter evolution can effectively enhance the detection performance of the network model for all levels of lotus after harvest. The data show that the improved algorithm has a better recognition effect for lotus seeds and meets the requirement for real-time detection, providing a technical reference for research on lotus quality classification.