Publications
Sort:
Issue
Accurate recognition of ginseng appearance quality based on improved YOLOv8n lightweighting
Transactions of the Chinese Society of Agricultural Engineering 2024, 40(24): 274-282
Published: 30 December 2024
Abstract PDF (1.6 MB) Collect
Downloads:8

Ginseng has been regarded as the top of conventional Chinese medicine. Its root can be widely used as an herbal medicine with a long history and multiple medicinal values; Thus, it is very crucial to monitor and manage ginseng quality. In this study, a lightweight CGC-YOLOv8 model was introduced into the detection system, in order to classify and assess the appearance quality of ginseng. Ginseng was harvested in Fusong County, Jilin Province, China. An experimental sample was then selected in the current year. Data enhancement techniques were used to simulate the various typical environments. In the basic network design of the model, the regular convolutional layer was replaced with two conditional convolutional layers (CondConv) in the backbone, according to the architecture of the YOLOv8 model. The appearance feature of input ginseng was highlighted to enhance the accuracy and efficiency of feature extraction. In addition, a thin neck combination (GSconv+VoVGSCSP) in the neck structure was applied to lighten the model. The coordinate Attention mechanism (CA) was incorporated to detect the head of the model. The performance was effectively improved without increasing the computational burden, particularly suitable for resource-limited devices. The CGC-YOLOv8 model was then optimized by ablation experiments. The better performance was achieved, with 85.70% precision, a recall of 91.23%, IoU=0.50 mean average precision (mAP50) of 94.69%, and IoU=0.95 mean average precision (mAP50-95)of 72.43%. Furthermore, the CGC-YOLOv8 improved the precision by 2.74 percentage points, recall by 4.64 percentage points, mAP50 by 3.71 percentage points, and mAP50-95 by 4.09 percentage points, respectively, compared with the YOLOv8n. The improved model outperformed the original one when detecting images without training. In addition, the number of parameters was reduced by 6.39% in the CGC-YOLOv8 model, compared with the original. The weight size was only 6.0 MB, fully meeting the lightweight condition and easy to deploy on mobile devices. A series of experiments were conducted to compare the effects of different attention mechanisms (CBAM, EMA, SE, SIMAM, and CA) on the improved YOLOv8n model. The application of the CA mechanism was also explored in the different feature layers (such as before SPPF, after SPPF, upsampling P4 layer, and small object detection layers P3, P4, and P5). The results demonstrated that the CA mechanism performed best when applied to the P5 layer. The recall and mean average precision were significantly enhanced in the improved model. The CGC-YOLOv8 model performed better than the conventional SSD EfficientDet and Yolo series models. Specifically, the precision of CGC-YOLOv8 was 85.20%. Secondly, the recall of CGC-YOLOv8 was 91.23%. The mean average precision (mAP50)was 94.69%. Finally, the mean average precision at the higher thresholds (mAP50-95) of the CGC-YOLOv8 was 72.43%. Therefore, the CGC-YOLOv8 model significantly outperformed the original and the rest, in terms of precision, recall, and mean average precision (mAP). Anyway, the CGC-YOLOv8 model can be expected to serve as a highly effective solution for intelligent detection of ginseng quality. The findings can also provide solid technical support for further advancements in the field.

Open Access Regular Issue
Research on Adaptive Optics Image Restoration Algorithm Based on Improved Joint Maximum a Posteriori Method
Photonic Sensors 2018, 8(1): 22-28
Published: 23 November 2017
Abstract PDF (447.1 KB) Collect
Downloads:47

In this paper, we propose a point spread function (PSF) reconstruction method and joint maximum a posteriori (JMAP) estimation method for the adaptive optics image restoration. Using the JMAP method as the basic principle, we establish the joint log likelihood function of multi-frame adaptive optics (AO) images based on the image Gaussian noise models. To begin with, combining the observed conditions and AO system characteristics, a predicted PSF model for the wavefront phase effect is developed; then, we build up iterative solution formulas of the AO image based on our proposed algorithm, addressing the implementation process of multi-frame AO images joint deconvolution method. We conduct a series of experiments on simulated and real degraded AO images to evaluate our proposed algorithm. Compared with the Wiener iterative blind deconvolution (Wiener-IBD) algorithm and Richardson-Lucy IBD algorithm, our algorithm has better restoration effects including higher peak signal-to-noise ratio (PSNR) and Laplacian sum (LS) value than the others. The research results have a certain application values for actual AO image restoration.

Total 2