Table 8 in An optimised YOLOv 4 deep learning model for efficient malarial cell detection in thin blood smear images

Table 8 Detection of infected RBC on Dataset B using the original YOLOv4 and modified models ModificationsModelPrecision (%)Recall rate (%)F1-score (%)mAP (%)Inference time (ms)B-FLOPSSize (MB)OriginalYOLOv461867281.43905.6559.57244.40Residual Block pruningYOLOv4-RC350956688.91695.5447.59242.40YOLOv4-RC450916585.20695.9851.21233.20YOLOv4-RC561937389.84731.4357.61222.10YOLOv4-RC3_459967490.70684.9337.35221.50YOLOv4-RC3_559937288.09690.5345.64220.40Backbone replacementYOLOv4- ResNet-50L54836576.95892.8037.33209.30YOLOv4-ResNet-50 M65817278.96905.3937.33209.30 B-FLOPS Billion floating-point operations, F1-SCoRE balance between precision and recall, mAP mean average precision表8 基于数据集B使用原始YOLOv4与改进模型对受感染红细胞（infected RBC）的检测结果 | 修改方式 | 模型 | 精确率（Precision (%)） | 召回率（Recall rate (%)） | F1分数（F1-score (%)） | 平均精度均值（mAP (%)） | 推理时延（Inference time (ms)） | 十亿次浮点运算量（B-FLOPS） | 模型体积（Size (MB)） | | --- | --- | --- | --- | --- | --- | --- | --- | --- | | 原始模型 | YOLOv4 | 61 | 86 | 72.81 | 43.90 | 55.65 | 59.57 | 244.40 | | 残差块剪枝 | YOLOv4-RC3 | 50 | 95 | 66.88 | 9.84 | 47.59 | 242.40 | 242.40 | | 残差块剪枝 | YOLOv4-RC4 | 50 | 91 | 65.85 | 69.58 | 51.21 | 233.20 | 233.20 | | 残差块剪枝 | YOLOv4-RC5 | 61 | 93 | 73.89 | 73.14 | 57.61 | 222.10 | 222.10 | | 残差块剪枝 | YOLOv4-RC3_4 | 59 | 99 | 67.40 | 68.49 | 37.35 | 221.50 | 221.50 | | 残差块剪枝 | YOLOv4-RC3_5 | 59 | 99 | 72.80 | 69.05 | 45.64 | 220.40 | 220.40 | | 主干网络替换 | YOLOv4-ResNet-50L | 54 | 83 | 65.76 | 89.28 | 37.33 | 209.30 | 209.30 | | 主干网络替换 | YOLOv4-ResNet-50M | 65 | 81 | 72.89 | 90.53 | 37.33 | 209.30 | 209.30 | 注：B-FLOPS指十亿次浮点运算（Billion floating-point operations）；F1分数为精确率与召回率的平衡度量；mAP为平均精度均值（mean average precision）