Precision agriculture-based orchard management relies heavily on the accurate delineation of tree canopies, especially for high-value crops like mangoes. Traditional GIS and remote sensing methods, such as Object-Based Imagery Analysis (OBIA), often face challenges due to overlapping canopies, complex tree structures, and varied light conditions. This study aims to enhance the accuracy of mango orchard mapping by developing a novel deep-learning approach that combines fine-tuned object detection and segmentation techniques. UAV imagery was collected over a 65-acre mango orchard in Multan, Pakistan, and processed into an RGB orthomosaic with a 3 cm ground sampling distance. The You Only Look Once (YOLOv7) framework was trained on an annotated dataset to detect individual mango trees. The resultant bounding boxes were used as prompts for the segment anything model (SAM) for precise delineation of canopy boundaries. Validation against ground truth data of 175 manually digitized trees showed a strong correlation ($R^2$ = 0.97), indicating high accuracy and minimal bias. The proposed method achieved a mean absolute percentage error (MAPE) of 4.94% and root mean square error (RMSE) of 80.23 sq ft against manually digitized tree canopies with an average size of 1290.14 sq ft. The proposed approach effectively addresses common issues such as inaccurate bounding boxes and over- or under-segmentation of tree canopies. The enhanced accuracy can substantially assist in various downstream tasks such as tree location mapping, canopy volume estimation, health monitoring, and crop yield estimation.

Mango (Mangifera Indica L.), a key horticultural crop, particularly in Pakistan, has been primarily studied locally using low- to medium-resolution satellite imagery, usually focusing on a particular phenological stage. The large canopy size, complex tree structure, and unique phenology of mango trees further accentuate intrinsic challenges posed by low-spatiotemporal-resolution data. The absence of mango-specific vegetation indices compounds the problem of accurate health classification and yield estimation at the tree level. To overcome these issues, this study utilizes high-resolution multi-spectral UAV imagery collected from two mango orchards in Multan, Pakistan, throughout the annual phenological cycle. It introduces MangiSpectra, an integrated two-staged framework based on Long Short-Term Memory (LSTM) networks. In the first stage, nine conventional and three mango-specific vegetation indices derived from UAV imagery were processed through fine-tuned LSTM networks to classify the health of individual mango trees. In the second stage, associated data such as the trees’ age, variety, canopy volume, height, and weather data were combined with predicted health classes for yield estimation through a decision tree algorithm. Three mango-specific indices, namely the Mango Tree Yellowness Index (MTYI), Weighted Yellowness Index (WYI), and Normalized Automatic Flowering Detection Index (NAFDI), were developed to measure the degree of canopy covered by flowers to enhance the robustness of the framework. In addition, a Cumulative Health Index (CHI) derived from imagery analysis after every flight is also proposed for proactive orchard management. MangiSpectra outperformed the comparative benchmarks of AdaBoost and Random Forest in health classification by achieving 93% accuracy and AUC scores of 0.85, 0.96, and 0.92 for the healthy, moderate and weak classes, respectively. Yield estimation accuracy was reasonable with $R^2=0.21$, and $\mathrm{RMSE}=50.18$. Results underscore MangiSpectra’s potential as a scalable precision agriculture tool for sustainable mango orchard management, which can be improved further by fine-tuning algorithms using ground-based spectrometry, IoT-based orchard monitoring systems, computer vision-based counting of fruit on control trees, and smartphone-based data collection and insight dissemination applications.