DENSE SMALL TARGET DETECTION ALGORITHM FOR FRESHWATER SNAILS BASED ON IMPROVED YOLOV8-OBB
-
摘要:
针对淡水螺分类加工场景中密集小目标检测存在的挑战, 文章提出了一种基于改进YOLOv8-OBB的淡水螺密集小目标检测算法。针对现有算法在复杂背景、目标个体小及类间特征差异小等场景下的性能不足, 文章通过两阶段创新策略优化模型: 首先, 基于SPDConv对P2层特征进行空间重构, 结合CSP与Omni-Kernel构建轻量级多尺度特征整合结构, 有效融合全局语义与局部细节信息, 提升密集小目标的特征表达能力; 其次, 提出改进的C2f-SREM模块, 通过Sobel边缘检测分支与四层卷积并行架构, 结合三重残差连接优化数据流传递, 强化模型对细微特征及边缘信息的捕捉能力。试验结果表明, 改进算法在自制淡水螺数据集上的mAP0.5达到80.6%, 较原YOLOv8n-OBB模型提升11.6%, 显著降低了漏检率与误检率。研究为淡水螺自动化分类加工提供了高效解决方案, 为密集小目标检测领域提供了新的技术参考, 推动水产品加工环节的智能化升级。
Abstract:In the integrated aquaculture system, a multi-species polyculture mode is commonly adopted, where freshwater economic species with ecological complementarity such as fish, crustaceans, and shellfish are co-cultivated in the same water body. To meet the differentiated demands for product specifications in the market, accurate sorting and processing according to biological species are required during the harvesting operation stage. This approach not only ensures the commercial value of various aquatic products and improves the efficiency at the sales end, but also optimize the management efficiency of the overall production and processing chain. In the classification and processing scenario of freshwater snail products, various snail species usually need to be accurately classified and graded for processing after fishing operations. The classification and detection of freshwater snail species are the basis for the automated processing of snail products, and it is of great significance in the industrialized cultivation, fishing, product processing, and classified sales of freshwater snails. Currently, machine vision technology based on deep learning is commonly applied to the classification and grading of agricultural products. However, in the classification operation link, the number of freshwater snails is usually huge, and as dense small targets, they are difficult to detect. Existing target detection algorithms still have deficiencies in perceiving dense small targets. Therefore, in response to the modernization needs of China's fishery industry, researching accurate and efficient detection methods for dense small target like freshwater snails is essential to promote automation in snail classification and processing. The development of automated aquaculture for freshwater snails is later than that of other aquatic organisms, with relatively few targeted automation and intelligence studies. Moreover, the algorithms described in relevant literature still have insufficient recognition effects for dense small targets of freshwater snails. In addition, different types of freshwater snails exhibit various shapes. When horizontal detection frames are used, a large amount of redundant information is included, leading to overlap significantly between frames. The use of Non-Maximum Suppression (NMS) may result in missed detections, which significantly impacts the model performance. This problem is particularly pronounced when freshwater snails are densely and overlappingly distributed, with subtle inter-class feature differences and complex backgrounds, their recognition performance is obviously insufficient. To effectively solve these problems, this paper innovatively proposes a dense small target detection algorithm for freshwater snails based on the improved YOLOv8-OBB algorithm. This algorithm processes the P2 feature layer through the introduction of SPDConv to obtain features rich in small target information, and fuses these features with the P3 layer. On this basis, the CSP and Omni-Kernel modules are combined for improved integration to obtain a new small target feature integration structure of COK, enhancing the network's perception ability for dense targets. The improved structure has increased the mAP0.5 index by 3.9%. Additionally,, an improved C2f-SREM module is proposed, incorporating parallel branches of SobelConv and additional convolution with a four-layer convolutional neural network and a triple residual connection architecture. This design greatly expands the global receptive field of the model and significantly enhances the context modeling ability, making the improved model more accurate in small target recognition. Compared with the original structure C2f, the improved module has increased the mAP0.5 index by 1.2%. From the perspective of the overall improved network model, the mAP0.5has increased by 11.6% compared to the original network, demonstrating obvious performance advantages. This research is of great significance for the development of the freshwater snail industry. In the industrialized cultivation of freshwater snails and the classification and grading processing of snail products after harvesting operations, the research results can provide reliable theoretical support, helping to promote the transformation and upgrading of the aquatic product processing industry such as freshwater snail classification and grading towards automation and intelligence, effectively improving industrial efficiency and increasing economic benefits.
-
-
![]()
图 1 数据集采集及分拣设备和采集过程示意图
a. 数据采集和分拣设备; b. 采集过程示意图
Figure 1. Data collection and sorting equipment and collection process diagram
a. data collection and sorting equipment; b. collection process diagram
![]()
图 2 自制淡水螺图像数据集和数据标注
a. 自制淡水螺图像数据集; b. 数据标注过程
Figure 2. Self-made freshwater snail image dataset and data annotation
a. self-made freshwater snail image dataset; b. data annotation process
![]()
图 4 Omni-Kernel模块和COK(CSP-Omnikernel)模块结构图
a. Omni-Kernel模块结构图; b. COK(CSP-Omnikernel)模块结构图
Figure 4. Omni-Kernel module and COK(CSP-Omnikernel) module architecture diagram
a. Omni-Kernel module architecture diagram; b. COK (CSP-Omnikernel) module architecture diagram
![]()
图 5 SREM模块和C2f_SREM模块结构图
a. SREM模块结构图; b. C2f_SREM模块结构图
Figure 5. SREM module and C2f_SREM module architecture diagram
a. SREM module architecture diagram; b. C2f_SREM module architecture diagram
![]()
图 7 淡水螺种类识别中模型改进前后的mAP0.5和box loss结果可视化
a. 模型mAP0.5对比图; b. 模型边界损失值对比图
Figure 7. Visualization of mAP0.5 and box loss results before and after model improvement in freshwater snail species recognition
a. Comparison chart of model mAP0.5; b. Comparison of Model Boundary Loss Values
![]()
图 8 改进前后检测结果可视化
a. 原图; b. 改进前; c. 改进后
Figure 8. Visualization of detection results before and after improvement
a. Original image; b. Before improvement; c.After improvement
表 1 评价指标释义
Table 1 Evaluation index interpretation
指标名称Indicator 具体含义Meaning 精确率 模型识别正类中实际为正类的比例 召回率 模型识别正类中占实际正类样本的比例 mAP0.5 网络在IOU阈值为0.5条件下的精度值 
下载: 导出CSV
表 2 深度学习的超参数配置
Table 2 Parameter configuration of deep learning
参数名称Parameter 具体配置Configuration 图像尺度 Input-shape 640×640 每批次样本数量 Batch-size 32 工作线程数 Num-workers 16 最小学习率 Mini learning rate 0.0001 最大学习率 Max learning rate 0.01 优化器种类 Optimizer SGD 训练轮数 Epochs 150
下载: 导出CSV
表 3 不同算法添加特征整合结构的对比结果
Table 3 Comparison results of different algorithms for adding feature integration structures
试验
Experiment模型
Model特征整
合结构
Feature
integration
structure精确率
Precisio
(%)召回率
Recall
(%)mAP0.5
Mean
Average
Precision
(0.5 %)1 YOLOv3-Tiny × 51.3 41.2 41.5 2 YOLOv3-Tiny √ 57.3 48.8 50.5 3 YOLOv5n × 61.8 66.1 65.7 4 YOLOv5n √ 70.3 66.5 71.4 5 YOLOv6n × 57 53.1 54.4 6 YOLOv6n √ 58.6 60.4 60 7 YOLOv8n × 62.9 69 68.7 8 YOLOv8n √ 70.2 69.6 72.6
下载: 导出CSV
表 4 不同算法添加改进C2f-SREM模块的对比结果
Table 4 Comparison results of different algorithms for adding feature integration structures
试验
Experiment模型
Model改进C2f-
SREM模块
Improved
C2f-SREM
module精确率
Precision
(%)召回率
Recall
(%)mAP0.5
Mean
Average
Precision
0.5 (%)1 YOLOv3-Tiny × 51.3 41.2 41.5 2 YOLOv3-Tiny √ 55.9 47.1 47.7 3 YOLOv5n × 61.8 66.1 65.7 4 YOLOv5n √ 68.2 66.7 68.9 5 YOLOv6n × 57 53.1 54.4 6 YOLOv6n √ 61.7 64.9 64.7 7 YOLOv8n × 62.9 69 68.7 8 YOLOv8n √ 67.7 70.9 69.9
下载: 导出CSV
表 5 C2f-SREM模块添加不同残差连接层数的对比结果
Table 5 Comparison results of different residual connection layers added to the C2f-SREM module
试验
Experiment残差连接层数
Rresidual connection layers精确率
Precision (%)召回率
Recall (%)mAP0.5
Mean Average Precision 0.5 (%)1 0层 67.4 70.4 70.1 2 1层 67.8 70.9 70.2 3 2层 66.3 69.5 69.9 4 3层 69.4 76.3 71.6 5 4层 66.8 69.2 69.2
下载: 导出CSV
表 6 整体改进网络模型消融试验
Table 6 Improved model ablation experiment
试验
Experiment调整上采样的引入输入层
Adjust the input layer for upsampling特征整合结构
Feature integration structureC2f-SREM Improved
C2f-SREM module精确率
Precision (%)召回率
Recall (%)mAP0.5 Mean
Average Precision
0.5 (%)1 × × × 62.9 69 68.7 2 √ × × 65.9 72 72.3 3 × √ × 70.2 69.6 72.6 4 × × √ 67.7 70.9 69.9 5 √ √ × 61.4 64.5 65.7 6 √ × √ 74.9 78.3 79.8 7 × √ √ 73.4 74.8 76.4 8 √ √ √ 74.3 80.3 80.6
下载: 导出CSV
表 7 数据集在不同算法中的检测结果对比
Table 7 Comparison of detection results of the dataset in different algorithms
试验
Experiment模型
Model精确率
Precision (%)召回率
Recall (%)mAP0.5
Mean Average Precision 0.5 (%)1 YOLOv7-tiny 58.8 62.4 60.4 2 YOLOv8n 62.9 69 68.7 3 YOLOv10n 61.0 64.8 62.9 4 YOLOv11n 61.9 66.6 64.7 5 Faster-RCNN 59.8 53.6 56.8 6 SSD 58.1 51.7 53.0 7 文献1 62.3 68.6 67.9 8 文献2 62.2 68.0 66.3 9 Ours 74.3 80.3 80.6
下载: 导出CSV
-
[1] 李道亮, 刘畅. 人工智能在水产养殖中研究应用分析与未来展望 [J]. 智慧农业, 2020, 2(3): 1-20.] Li D L, Liu C. Recent advances and future outlook for artificial intelligence in aquaculture [J]. Smart Agriculture, 2020, 2(3): 1-20. [
[2] 李威锋, 宋烨佳, 文衍红, 等. 我国淡水螺产业现状调查 [J]. 养殖与饲料, 2020, 19(11): 131-138.] Li W F, Song Y J, Wen Y H, et al. Investigation on the current situation of the freshwater snail industry in China [J]. Animals Breeding and Feed, 2020, 19(11): 131-138. [
[3] 刘世晶, 李国栋, 刘晃, 等. 中国水产养殖装备发展现状 [J]. 水产学报, 2023, 47(11): 119615.] Liu S J, Li G D, Liu H, et al. Current development status of aquaculture equipment in China [J]. Journal of Fisheries of China, 2023, 47(11): 119615. [
[4] 段延娥, 李道亮, 李振波, 等. 基于计算机视觉的水产动物视觉特征测量研究综述 [J]. 农业工程学报, 2015, 31(15): 1-11.] Duan Y E, Li D L, Li Z B, et al. Review on visual characteristic measurement research of aquatic animals based on computer vision [J]. Transactions of the Chinese Society of Agricultural Engineering, 2015, 31(15): 1-11. [
[5] 靳学萌, 梁西银, 邓鹏飞. 基于改进YOLOv10的轻量级黄花菜分级检测模型 [J]. 智慧农业, 2024, 6(5): 108-118.] Jin X M, Liang X Y, Deng P F. Lightweight daylily grading and detection model based on improved YOLOv10 [J]. Smart Agriculture, 2024, 6(5): 108-118. [
[6] 陈科, 周勇, 薛明洋, 等. 基于机器视觉和改进YOLOv5s的鲫病害轻量级无损检测模型 [J]. 水生生物学报, 2024, 48(7): 1141-1148.] Chen K, Zhou Y, Xue M Y, et al. Lightweight nondestructive detection model of crucian carp disease based on machine vision and improved YOLOv5s [J]. Acta Hydrobiologica Sinica, 2024, 48(7): 1141-1148. [
[7] 冯怡然, 张馨丹, 陶学恒. 基于改进的Faster R-CNN贝类识别检测 [J]. 大连工业大学学报, 2022, 41(3): 208-213.] Feng Y R, Zhang X D, Tao X H. Recognition and detection of shellfish based on improved Faster R-CNN [J]. Journal of Dalian Polytechnic University, 2022, 41(3): 208-213. [
[8] 朱明, 李梦珂, 万鹏, 等. 基于鱼体特征点检测的淡水鱼种类识别 [J]. 农业工程学报, 2023, 39(11): 155-164.] Zhu M, Li M K, Wan P, et al. Identification of freshwater fish species based on fish feature point detection [J]. Transactions of the Chinese Society of Agricultural Engineering, 2023, 39(11): 155-164. [
[9] 施亮, 熊春蓉, 刘毛毛, 等. 基于深度学习技术的湖北钉螺视觉智能识别模型的建立 [J]. 中国血吸虫病防治杂志, 2021, 33(5): 445-451.] Shi L, Xiong C R, Liu M M, et al. Establishment of a deep learning-based visual model for intelligent recognition of Oncomelania hupensis [J]. Chinese Journal of Schistosomiasis Control, 2021, 33(5): 445-451. [
[10] Fang J, Meng J, Liu X, et al. Single-target detection of Oncomelania hupensis based on improved YOLOv5s [J]. Frontiers in Bioengineering and Biotechnology, 2022(10): 861079.
[11] 袁红春, 白宝来, 陶磊. 基于改进YOLOv5_OBB的中华绒螯蟹旋转目标检测 [J]. 福建农林大学学报(自然科学版), 2024, 53(2): 284-288.] Yuan H C, Bai B L, Tao L. Rotating target detection of Chinese Eriocheir sinernsis based on the improved YOLOv5_OBB [J]. Journal of Fujian Agriculture and Forestry University ( Natural Science Edition), 2024, 53(2): 284-288. [
[12] Cui J, Zhang X, Xiong F, et al. Pathological myopia image recognition strategy based on data augmentation and model fusion [J]. Journal of Healthcare Engineering, 2021(1): 5549779.
[13] Krizhevsky A, Sutskever I, Hinton G E. ImageNet classification with deep convolutional neural networks [J]. Communications of the ACM, 2017, 60(6): 84-90. doi: 10.1145/3065386
[14] Singha S, Aydin B. Automated drone detection using YOLOv4 [J]. Drones, 2021, 5(3): 95. doi: 10.3390/drones5030095
[15] Redmon J, Divvala S, Girshick R, et al. You only look once: unified, real-time object detection [C]//2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). June 27-30, 2016. Las Vegas, NV, USA. IEEE, 2016: 779-788.
[16] 牛为华, 郭迅. 基于改进YOLOv8的船舰遥感图像旋转目标检测算法 [J]. 图学学报, 2024, 45(4): 726-735.] Niu W H, Guo X. Rotating target detection algorithm in ship remote sensing images based on YOLOv8 [J]. Journal of Graphics, 2024, 45(4): 726-735. [
[17] 李丽, 卢世博, 任浩, 等. 基于改进YOLOv5的复杂环境下桑树枝干识别定位方法 [J]. 农业机械学报, 2024, 55(2): 249-257.] Li L, Lu S B, Ren H, et al. Mulberry branch identification and location method based on improved YOLO v5 in complex environment [J]. Transactions ofthe Chinese Society for Agricultural Machinery, 2024, 55(2): 249-257. [
[18] 罗友璐, 潘勇浩, 夏顺兴, 等. 基于改进YOLOv8的苹果叶病害轻量化检测算法 [J]. 智慧农业, 2024, 6(5): 128-138.] Luo Y L, Pan Y H, Xia S X, et al. Lightweight apple leaf disease detection algorithm based on improved YOLOv8 [J]. Smart Agriculture, 2024, 6(5): 128-138. [
[19] Cui Y, Ren W, Knoll A. Omni-Kernel network for image restoration [J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2024, 38(2): 1426-1434. doi: 10.1609/aaai.v38i2.27907
[20] Fu J, Liu J, Tian H, et al. Dual attention network for scene segmentation [C]//2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). June 15-20, 2019. Long Beach, CA, USA. IEEE, 2019: 3141-3149.
[21] Zhou S , Pan Y, Spectrum attention mechanism for time series classification [C]// 2021 IEEE 10th Data Driven Control and Learning Systems Conference (DDCLS). May 14-16, 2021. Suzhou, China. IEEE, 2021: 339-34.
[22] 马常昊, 胡文惠, 钟海超, 等. 融合Sobel算子的SAR图像结构优化方法 [J]. 探测与控制学报, 2024, 46(2): 119-124.] Ma C H, Hu W H, Zhong H C, et al. SAR image structure optimization method using Sobel operator fusion [J]. Journal of Detection & Control, 2024, 46(2): 119-124. [
[23] Farhadi A, Redmon J. Yolov3: An incremental improvement [C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition(CVPR). Berlin/Heidelberg, Germany: Springer, 2018, 1804: 1-6.
[24] Zhu L, Geng X , Li Z, et al. Improving YOLOv5 with attention mechanism for detecting boulders from planetary Images [J]. Remote Sensing, 2021, 13(18): 3776.
[25] Hussain M. YOLO-v1 to YOLO-v8, the rise of YOLO and its complementary nature toward digital manufacturing and industrial defect detection [J]. Machines, 2023, 11(7): 677. doi: 10.3390/machines11070677
[26] 周秀珊, 文露婷, 介百飞, 等. 改进YOLOv11的水面膨化饲料颗粒图像实时检测算法 [J]. 智慧农业, 2024, 6(6): 155-167.] Zhou X S, Wen L T, Jie B F, et al. Real-time detection algorithm of expanded feed image on the water surface based on improved YOLOv11 [J]. Smart Agriculture, 2024, 6(6): 155-167. [
[27] 王泽宇, 徐慧英, 朱信忠, 等. 基于YOLOv8改进的密集行人检测算法: MER-YOLO [J]. 计算机工程与科学, 2024, 46(6): 1050-1062.] Wang Z Y, Xu H Y, Zzhu X Z, et al. An improved dense pedestrian detection algorithm based on YOLOv8: MER-YOLO [J]. Computer Engineering & Science, 2024, 46(6): 1050-1062. [
[28] Li Y, Huang Q, Pei X, et al. Cross-layer attention network for small object detection in remote sensing imagery [J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2021(14): 2148-2161.
[29] 宋晓茹, 刘康, 高嵩, 等. 复杂战场环境下改进YOLOv5军事目标识别算法研究 [J]. 兵工学报, 2024, 45(3): 934-947.] Song X R, Liu K, Gao S, et al. Research on improved YOLOv5-based military target recognition algorithm used in complex battlefield environment [J]. Acta Armamentarii, 2024, 45(3): 934-947. [
[30] 许迪, 张淑卿, 葛超. 面向复杂环境的YOLOv8安全装备检测 [J]. 电子测量技术, 2024, 47(7): 121-129.] Xu D, Zhang S Q, Ge C. YOLOv8security equipment inspection for complex environment [J]. Electronic Measurement Technology, 2024, 47(7): 121-129. [
-
期刊类型引用(19)
1. 魏正玲，罗凯，邵仙萍，郜卫华，吴成龙，赵成民. 菜籽粕替代鱼粉对青鱼生长、抗氧化能力和肠道健康的影响. 动物营养学报. 2024(06): 3876-3891 . 
百度学术
2. 陈龙，刘小玲，阚画春，王维，郑曙峰，李淑英，陈敏，陈超，徐道青. 棉粕饲料化高效利用分析. 安徽农学通报. 2024(12): 68-73 . 百度学术
3. 陈倪慧，杨霞，孙珊珊，雷西汀，邵仙萍，吴成龙，叶金云. 发酵植物蛋白复配替代豆粕对青鱼生长、免疫、抗氧化及肠道菌群结构的影响. 动物营养学报. 2023(04): 2478-2489 . 百度学术
4. 付豪，梁艺馨，牟希东，刘超，杨叶欣，刘奕，宋红梅. 发酵南极磷虾粉对锦鲤生长性能、血清生化指标、肠道结构及菌群的影响. 广东海洋大学学报. 2023(05): 34-42 . 百度学术
5. 王冰柯，张芹，王延晖. 饲料蛋白质水平对水产动物生长及抗氧化的影响. 河南水产. 2021(04): 3-5 . 百度学术
6. 苏美珍，张祎桐，王浩，吕利群. 甲苯咪唑在异育银鲫体内代谢及肝毒性研究. 水产科学. 2020(05): 710-718 . 百度学术
7. 申建飞，陈铭灿，刘泓宇，谭北平，董晓慧，杨奇慧，迟淑艳，章双，王华朗. 浓缩棉籽蛋白替代鱼粉对卵形鲳鲹幼鱼生长性能、血清生化指标、肝脏抗氧化指标及胃肠道蛋白酶活性的影响. 动物营养学报. 2019(02): 746-756 . 百度学术
8. 周凡，贝亦江，李诗言，王鼎南，丁雪燕，周万华. 不同蛋氨酸剂型对中华鳖生长性能和肝脏基因表达的影响. 水产科技情报. 2019(02): 73-80 . 百度学术
9. 孙金秋，徐莞媛，高焕，阎斌伦. 中华绒螯蟹营养生理学研究进展. 水产科技情报. 2018(04): 197-202 . 百度学术
10. 张海涛，蒲雪松，杨奇慧，谭北平，董晓慧，迟淑艳，刘泓宇，章双. 高蛋白棉粕替代鱼粉对凡纳滨对虾生长性能、非特异性免疫指标及抗病力的影响. 广东海洋大学学报. 2018(04): 20-26 . 百度学术
11. 于海馨，徐东坡，周游，游磊，杨彦平，刘凯. 中华绒螯蟹生殖群体生物学特征及消化酶活力研究. 上海海洋大学学报. 2018(06): 844-854 . 百度学术
12. 何娇娇，王萍，冯建，娄宇栋，党慧，邓蓉. 玉米蛋白粉替代鱼粉对大黄鱼生长、血清生化指标及肝脏组织学的影响. 水生生物学报. 2017(03): 506-515 . 本站查看
13. 崔燕燕，张南南，马倩倩，陈晴，沈振华，杜震宇，陈立侨. 四种植物蛋白对中华绒螯蟹幼蟹生长性能、氨基酸沉积率和抗氧化酶活性的影响. 水生生物学报. 2017(01): 146-154 . 本站查看
14. 姚清华，颜孙安，郭清雄，王宝珍，陆鹏，林虬. 羽毛肽粉对瓦氏黄颡鱼稚鱼生长、体成分和消化酶活性的影响. 福建农业学报. 2016(06): 580-585 . 百度学术
15. 向枭，周兴华，曾本和，罗莉，文华. 蛋氨酸水平对吉富罗非鱼肌肉氨基酸组成及血清抗氧化能力的影响. 水产学报. 2016(09): 1359-1367 . 百度学术
16. 杨霞，叶金云，张易祥，吴成龙，刘沛，王伟. 普通棉籽粕和发酵棉籽粕替代鱼粉对中华绒螯蟹幼蟹生长性能、体成分及肝胰腺消化酶活性的影响. 动物营养学报. 2014(03): 683-693 . 百度学术
17. 徐奕晴，祝向阳，颜培实. 生物发酵饲料对中华绒螯蟹幼蟹生长、饲料利用及抗氧化酶活性的影响. 江苏农业科学. 2014(07): 237-240 . 百度学术
18. 严全根，朱晓鸣，杨云霞，韩冬，金俊琰，解绶启，李钟杰. 饲料中棉粕替代鱼粉蛋白对草鱼的生长、血液生理指标和鱼体组成的影响. 水生生物学报. 2014(02): 362-369 . 本站查看
19. 杨霞，叶金云，周志金，张易祥，吴成龙，明建华. 中华绒螯蟹幼蟹对亮氨酸和异亮氨酸的需要量. 水生生物学报. 2014(06): 1062-1070 . 本站查看
其他类型引用(29)
计量
- 文章访问数: 14
- HTML全文浏览量: 3
- PDF下载量: 2
- 被引次数: 48