【摘要】 为提高前臀鮡（Pareuchiloglanis anteanalis）表型数据获取效率并降低人工测量误差，构建基于关键点检测的表型自动测量与体质量预测方法。采集前臀鮡腹部视角、侧面视角与背部视角图像共951张，采用COCO格式完成关键点标注，并基于关键点坐标计算体长、体高、眼间距等传统体尺及框架距离指标；在相同训练配置下对RTMPOSE、LiteHRNet、YOLOv12n-pose与YOLOv8n-pose进行对比评估。结果显示，YOLOv8npose在测试集上的精确率为94.70%，表型测量平均相对误差（mean relative error,MRE）为5.45%，多数表型相对误差控制在10%以内；进一步结合相关性与共线性分析筛选胸鳍基部起点间距离（X₅）、胸鳍基部起点到腹鳍基部右端起点距离（X₇）、体高(X₁₀)、头长(X₁₂)和眼间距(X₁₄)等表型指标建立体质量多元回归模型，测试集决定系数R²为0.97。结果表明，该方法可实现前臀鮡表型的自动化测量与体质量的定量估算。更多还原

【Abstract】 A method of automatically measuring the phenotype and predicting the body weight of Pareuchiloglanis anteanalis based on the keypoint detection was developed to improve the efficiency of obtaining the phenotypic data of P. anteanalis and reduce errors caused by manual measurements. 951 images were collected from the ventral,lateral,and dorsal views of P. anteanalis.Keypoints were annotated in the COCO format.Traditional body size including body length,body height,and interorbital distance and indexes of frame distance were calculated based on the coordinates of keypoints.RTMPOSE,LiteHRNet,YOLOv12n-pose,and YOLOv8n-pose were comparatively evaluated under the same configuration of training,and YOLOv8n-pose was determined to be used for the automatic measurement of phenotypic parameters.Results showed that YOLOv8n-pose had a precision of 94.70% on the test set,with mean relative error（MRE） of 5.45% for phenotypic measurements,and the relative error of most phenotypes was controlled within 10%.Key indexes of phenotype including the distance between pectoral-fin origins（X₅）,the distance from the pectoral-fin base to the right pelvic-fin base（X₇）,body height(X₁₀),head length(X₁₂),and interorbital distance(X₁₄),were further selected by combining correlation and collinearity analysis to establish a multiple regression model for predicting body-weight,yielding an R² of 0.97 on the test set.It is indicated that the proposed method can achieve automated measurement of phenotype and quantitative estimation of body weight of P. anteanalis. It will provide data support for monitoring growth and evaluating selective breeding of small sisorid freshwater fishes.更多还原