

















摘要:
為實(shí)現(xiàn)不同等級(jí)的龍井茶鮮葉在線分級(jí),提出一種基于機(jī)器視覺的茶鮮葉在線檢測(cè)與分級(jí)方法,設(shè)計(jì)并制造一套智能分級(jí)裝置。通過添加坐標(biāo)注意力機(jī)制、引入空洞空間卷積池化金字塔和改進(jìn)特征融合網(wǎng)絡(luò)對(duì)YOLOv5s進(jìn)行優(yōu)化得到Y(jié)OLOv5s—CAB,識(shí)別的平均精度均值為90.4%,召回率為87.8%。在茶鮮葉分級(jí)裝置上進(jìn)行試驗(yàn),結(jié)果表明:確定最佳參數(shù)茶鮮葉下落速度和傳送帶速度分別為2.08g/s、150.00mm/s時(shí),識(shí)別的準(zhǔn)確率達(dá)95.58%,驗(yàn)證裝置的可行性與可靠性,為茶鮮葉的智能化分級(jí)提供技術(shù)支撐。
關(guān)鍵詞:茶鮮葉;深度學(xué)習(xí);實(shí)時(shí)檢測(cè);智能分級(jí)
中圖分類號(hào):TP391.41; TS272.3
文獻(xiàn)標(biāo)識(shí)碼:A
文章編號(hào):2095-5553 (2025) 03-0139-07
收稿日期:2023年9月21日" 修回日期:2023年12月14日*
基金項(xiàng)目:浙江省食品物流裝備技術(shù)研究重點(diǎn)實(shí)驗(yàn)室開放基金(KF2022003yb)
第一作者:吳堅(jiān),男,1965年生,杭州人,碩士,教授;研究方向?yàn)槲锪餮b備、數(shù)控技術(shù)。E-mail: wujian@zust.edu.cn
Design and experiment of an intelligent fresh tea leaves grading device
Wu Jian1," 2, Ye Mengyan1, Zhang Tongfeng3
(1. Zhejiang University of Science and Technology, Hangzhou, 310023, China;
2. Zhejiang Key Laboratory of Food Logistics Equipment and Technology, Hangzhou, 310023, China;
3. Zhejiang Wason Cold Chain Technology Co., Ltd., Hangzhou, 310023, China)
Abstract:
To realize the online grading of Longjing tea fresh leaves of different quality levels, this study proposes a machine vision-based method for online detection and grading of fresh tea leaves, and designs and manufactures an intelligent grading device. By adding a coordinate attention mechanism, incorporating dilated spatial convolution pooling pyramid, and improving the feature fusion network, YOLOv5s is optimized into YOLOv5s—CAB. The average precision for recognition is 90.4%, and the recall rate is 87.8%. Experiments conducted on the fresh tea leaves grading device show that when the optimal parameters for the falling speed of the leaves and the conveyor belt speed are set at 2.08 g/s and 150.00mm/s, respectively, the recognition precision reaches 95.58%, which verifies the feasibility and reliability of the device and provides technical support for the intelligent grading of fresh tea leaves.
Keywords:
fresh tea leaves; deep learning; real-time detection; intelligent grading
0 引言
茶葉具有悠久歷史和豐富文化內(nèi)涵,我國茶葉有紅茶、綠茶、烏龍茶、黃茶、白茶和黑茶六大類[1]。在茶鮮葉采摘中,由于茶農(nóng)人員雜及采茶機(jī)本身固有屬性特點(diǎn),導(dǎo)致茶鮮葉間存在較大的差異,茶鮮葉的長度不一、老嫩不均、凈度不高等各種因素影響茶葉品質(zhì)。傳統(tǒng)的茶鮮葉分級(jí)為振動(dòng)式[2]、滾篩式[3]和風(fēng)力分選[4],主要利用茶鮮葉大小形狀進(jìn)行分類,分級(jí)效果差異較大。因此,為提高茶葉品質(zhì),研究茶鮮葉的智能化分級(jí)具有重要意義。
隨著機(jī)器視覺技術(shù)的發(fā)展與應(yīng)用,越來越多學(xué)者將其應(yīng)用于農(nóng)產(chǎn)品分級(jí)領(lǐng)域[5, 6],在茶葉的檢測(cè)與分級(jí)中也取得一定成果。……