Figures (10)  Tables (5)

    • Figure 1. 

      Display of three diseased and healthy leaves. (a) Measles. (b) Wilt. (c) Rotten holes. (d) Healthy.

    • Figure 2. 

      Flowchart of the improved YOLOv11n algorithm.

    • Figure 3. 

      C2f-DCN module structure diagram.

    • Figure 4. 

      Normalized confusion matrix diagram.

    • Figure 5. 

      Different models for disease detection under different background conditions. (a) Original image. (b) Against a strong light background. (c) Against a shaded background.

    • Figure 6. 

      Feature map visualization.

    • Figure 7. 

      Model detection of grape leaf diseases (a) Fusarium wilt, (b) measles, and (c) rotten holes.

    • Figure 8. 

      P and R curves of three types of object detection algorithms.

    • Figure 9. 

      Precision recall curves of the detection algorithm for three types of grape leaf diseases.

    • Figure 10. 

      Graphs generated by four algorithms for the detection of grape leaf diseases. (a) YOLOv11n. (b) YOLOv11n-DPPA. (c) DCD-YOLO. (d) DCD-YOLOs.

    • Disease category Train set Valid set Test set Combined
      Measles 1,333 167 167 1,667
      Wilt 1,333 167 167 1,667
      Rotten holes 1,334 166 166 1,666
      Total 4,000 500 500 5,000

      Table 1. 

      Dataset of grape leaf diseases.

    • DPPA DFF C2f-DCN P (%) R (%) mAP (%) Number of parameters (M) Model size (MB) Inference speed (ms) Frames per second (fps)
      — — — 89.6 90.9 94.5 2.58 5.23 3.5 285.71
      √ — — 90.5 90.2 94.2 4.65 9.64 4.8 208.33
      — √ — 86.2 92.1 94.1 17.67 35.54 11.6 90.91
      — — √ 91.6 90.7 95.0 2.77 5.57 3.4 294.12
      √ — √ 92.3 90.1 95.3 5.07 9.98 4.8 208.33
      YOLOv11s
      √ — √ 92.0 91.9 95.7 14.68 28.36 9.8 128.21

      Table 2. 

      Results of ablation trials.

    • Model Wilt (%) Measles (%) Rotten holes (%) mAP (%) Inference speed (ms) Frames per second (fps)
      YOLOv11n 0.981 0.870 0.985 0.945 3.5 285.71
      YOLOv11n-DPPA 0.978 0.865 0.983 0.942 4.8 208.33
      DCD-YOLO 0.983 0.899 0.976 0.953 4.8 208.33
      DCD-YOLOs 0.970 0.919 0.982 0.957 9.8 128.21
      YOLOv11n-DFF 0.968 0.876 0.978 0.941 11.6 90.91
      YOLOv8s 0.981 0.906 0.984 0.957 6.7 149.25
      YOLOv8l 0.986 0.883 0.976 0.948 25.1 39.84
      YOLOv12 0.977 0.775 0.978 0.910 5.9 169.50

      Table 3. 

      Comparison of average accuracy detection results of different networks on the test set.

    • Rounds P (%) R (%) mAP (%) Inference speed (ms)
      100 0.838 0.896 0.922 5.0
      150 0.861 0.899 0.929 4.7
      200 0.911 0.878 0.943 5.0
      250 0.923 0.901 0.953 4.8
      300 0.908 0.906 0.941 4.9

      Table 4. 

      Performance detection at different rounds.

    • Model P (%) R (%) mAP (%) Inference speed (ms)
      DCD-YOLOs 0.920 0.919 0.957 9.8
      DCD-YOLO 0.923 0.901 0.953 4.8
      YOLOv11n-DPPA 0.905 0.902 0.942 4.8
      YOLOv11n 0.896 0.909 0.945 3.5
      YOLOv11n-DFF 0.862 0.921 0.941 11.6

      Table 5. 

      Detection results of different models.