DySPAM-Net dataset

<p dir="ltr">This dataset contains field-collected mango (Mangifera indica L.) canopy images acquired for the development and evaluation of DySPAM-Net (Dynamic Sequential Phenology-Aware Multimodal Network), a deep-learning framework for early-stage mango yield prediction.</p><p dir="ltr">The dataset was collected from a commercial mango orchard in Gujarat, India, using smartphone-based field imaging. Twenty-five mango trees were observed longitudinally across four weekly phenological stages: (1) Peak Flowering, (2) Flower Drop / Fruit Set, (3) Small Fruit Development, and (4) Mature Fruit.

To obtain a more representative view of each tree canopy, images were acquired from four cardinal directions: North, South, East, and West.</p><p dir="ltr">A total of 292 field images were captured, of which 284 images passed quality screening and were retained for analysis. The quality-controlled dataset consists of 80 images from Week 1, 81 images from Week 2, 48 images from Week 3, and 75 images from Week 4. The images were subsequently used for image preprocessing, spatial feature extraction, phenology-aware feature encoding, temporal learning, developmental-stage classification, and mango yield prediction experiments.</p><p dir="ltr">The dataset supports research in computer vision, precision agriculture, crop phenology analysis, deep learning, temporal modelling, and early-stage fruit yield prediction.