基于改進(jìn)YOLOv8n的茶葉嫩稍檢測(cè)方法
農(nóng)業(yè)工程學(xué)報(bào)
頁(yè)數(shù): 10 2024-06-13
摘要: 針對(duì)名優(yōu)茶智能采摘中茶葉嫩梢識(shí)別精度不足的問(wèn)題,該研究對(duì)YOLOv8n模型進(jìn)行優(yōu)化。首先,在主干網(wǎng)絡(luò)中引入動(dòng)態(tài)蛇形卷積(dynamic snake convolution,DSConv),增強(qiáng)模型對(duì)茶葉嫩梢形狀信息的捕捉能力;其次,將頸部的路徑聚合網(wǎng)絡(luò)(path aggregation network,PANet)替換為加權(quán)雙向特征金字塔網(wǎng)絡(luò)(bi-directional fe...