Scholarlink

<p>Due to the reliance of deep learning on high-quality datasets and the uncertainty regarding the reproductive period segmentation and image annotation of the <i>Pleurotus citrinopileatus</i> maturity dataset, this study aims to enhance the accuracy of dataset annotation by leveraging the color change patterns of <i>Pleurotus citrinopileatus</i>. First, reproductive period division and color feature analysis were conducted. Specifically, considering the influence of the cultivation environment, a <i>Pleurotus citrinopileatus</i> color feature database was constructed using data from different environmental batches. Second, the growth process of <i>Pleurotus citrinopileatus</i> fruiting bodies was divided into four stages: primordium stage, young mushroom stage, forming stage, and maturation stage. Third, color pattern verification revealed that through HSV color space analysis, the main hue exhibited periodic changes. Experimental data showed that when the main hue exceeded 28°, it entered the next growth stage, decreasing to 23° before rising again, which validated the division of the four reproductive periods from primordium to maturity. Next, feature extraction and model construction were performed. Statistical measures such as mean, variance, and peak values of the H, S, and V channels for each sample were calculated to build a Gradient Boosting Tree (GBT) classification model based on color features. A new dataset was generated by adjusting expert annotations based on model validation. Finally, the experimental results of dataset quality were annotated and performance improvements were verified. Using the YOLO12 model for image feature learning of <i>Pleurotus citrinopileatus</i>, training and testing were conducted with both the original and improved datasets. The improved dataset(0.935 mAP@0.5 vs. 0.91 mAP@0.5), representing a 2.5 percentage point increase. On a test set of 417 images, the accuracy of the improved dataset reached 94.5%, compared to 92% for the original dataset, marking a 2.5 percentage point improvement. In conclusion, analyzing the color features of <i>Pleurotus citrinopileatus</i> effectively guides the division of maturity datasets, thereby enhancing the performance of deep learning-based <i>Pleurotus citrinopileatus</i> maturity recognition models.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Analysis of color patterns during the growth period of pleurotus citrinopileatus and optimization method for deep learning dataset

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月24日

查看全文
  • Jun Yu,
  • Changshou Luo,
  • Qingfeng Wei,
  • Yang Lu,
  • Yaming Zheng,
  • Rupeng Luan

摘要

Due to the reliance of deep learning on high-quality datasets and the uncertainty regarding the reproductive period segmentation and image annotation of the Pleurotus citrinopileatus maturity dataset, this study aims to enhance the accuracy of dataset annotation by leveraging the color change patterns of Pleurotus citrinopileatus. First, reproductive period division and color feature analysis were conducted. Specifically, considering the influence of the cultivation environment, a Pleurotus citrinopileatus color feature database was constructed using data from different environmental batches. Second, the growth process of Pleurotus citrinopileatus fruiting bodies was divided into four stages: primordium stage, young mushroom stage, forming stage, and maturation stage. Third, color pattern verification revealed that through HSV color space analysis, the main hue exhibited periodic changes. Experimental data showed that when the main hue exceeded 28°, it entered the next growth stage, decreasing to 23° before rising again, which validated the division of the four reproductive periods from primordium to maturity. Next, feature extraction and model construction were performed. Statistical measures such as mean, variance, and peak values of the H, S, and V channels for each sample were calculated to build a Gradient Boosting Tree (GBT) classification model based on color features. A new dataset was generated by adjusting expert annotations based on model validation. Finally, the experimental results of dataset quality were annotated and performance improvements were verified. Using the YOLO12 model for image feature learning of Pleurotus citrinopileatus, training and testing were conducted with both the original and improved datasets. The improved dataset(0.935 mAP@0.5 vs. 0.91 mAP@0.5), representing a 2.5 percentage point increase. On a test set of 417 images, the accuracy of the improved dataset reached 94.5%, compared to 92% for the original dataset, marking a 2.5 percentage point improvement. In conclusion, analyzing the color features of Pleurotus citrinopileatus effectively guides the division of maturity datasets, thereby enhancing the performance of deep learning-based Pleurotus citrinopileatus maturity recognition models.