Fig. 1

Schematic diagram of the CaneUNet-W model structure"

Table 1

Model structure parameter settings (example for input image size 512×512)"

Table 2

Drought classification standards"

Table 3

Regional crossover experimental design plan"

Fig. 2

Performance evaluation results of the model in the three-region cross-validation experiments. Evaluation metrics include IoU, mIoU, F1-score, Precision, Recall, and Kappa coefficient. Each experiment used data from two regions to train the model and tested it in the remaining regions. Subplots: Left (Grouped bar charts) show the scores of each metric; Right (ROC curves) show the classification performance and AUC value."

Table 4

Model performance evaluation metrics in different test areas"

Fig. 3

Study area distribution and experimental performance comparison: The geographical locations and model performance of the three experimental areas are marked. Area A performed best (mIoU=0.955, Kappa=0.954), Area B performed relatively poorly (mIoU=0.882, Kappa=0.874), and Area C, with its flat terrain and later data collection time, had slightly lower mIoU and Kappa. The figure visually illustrates the impact of topographic and phenological differences on the model's recognition accuracy."

Fig. 4

Visual example of sugarcane field segmentation and drought level prediction in a typical test area. The left column shows the original UAV imagery, the middle column shows the predicted drought levels, and the right column shows the overlay results, which intuitively demonstrate the model's accurate identification of sugarcane field edges and the objective and refined characterization of drought spatial conditions using the NDVI physical index."

Table 5

Performance comparison of different models in sugarcane field recognition tasks"

Fig. 5

compares the CaneUNet-W model with other models on various metrics and evaluates the overall performance of the three models. Evaluation metrics include IoU, mIoU, F1-score, Precision, Recall, and Kappa coefficient. The left figure shows the specific values of the three models on each metric and the overall performance differences; The right figure shows the ROC curves of the three models' classification performance and AUC values."

Fig. 6

Comparison of model performance evaluation and ablation experiment results. Performance analysis of five ablation models of the improved U-Net model on various evaluation metrics (IoU, mIoU, F1-score, Precision, Recall, Kappa). The left figure (Violin-box-means plot) shows the value distribution, dispersion, and mean comparison of each model on all metrics; The right figure (ROC curve plot) shows the classification performance of the five models and their corresponding AUC values."

Table 6

Comparison of ablation experiment results"