IBSA_Net: A Network for Tomato Leaf Disease Identification Based on Transfer Learning with Small Samples

Tomatoes are a crop of significant economic importance, and disease during growth poses substantial threat to yield quality. In this paper, we propose IBSA_Net, tomato leaf recognition network that employs transfer learning small sample data, while introducing the Shuffle Attention mechanism enhance feature representation. The model is optimized by employing IBMax module increase receptive field adding HardSwish function ConvBN layer improve stability speed. To address challenge poor generalization models trained on public datasets real environment datasets, developed an improved PlantDoc++ dataset utilized pre-train PDDA PlantVillage datasets. results indicate after pre-training dataset, IBSA_Net achieved test accuracy 0.946 with average precision, recall, F1-score 0.942, 0.944, 0.943, respectively. Additionally, effectiveness in other crops verified. This study provides dependable effective method for recognizing diseases agricultural production environments, potential application crops.