Hyperspectral Classification of Frost Damage Stress in Tomato Plants Based on Few-Shot Learning

Early detection and diagnosis of crop anomalies is crucial for enhancing yield quality. Recently, the combination machine learning deep with hyperspectral images has significantly improved efficiency detection. However, acquiring a large amount properly annotated data on stressed crops requires extensive biochemical experiments specialized knowledge. This limitation poses challenge to construction large-scale datasets stress analysis. Meta-learning approach that capable learn can achieve high accuracy limited training samples. In this paper, we introduce meta-learning imaging first time. addition, gathered 88 drought-stressed tomato plants 68 freeze-stressed plants. The related drought serve as source domain, while frost damage target domain. Due difficulty obtaining domain from real-world testing scenarios, only are used model training. results indicated meta-learning, minimum eight samples, achieved 69.57%, precision 59.29%, recall 66.32% F1-score 62.61% classifying severity stress, surpassing other methods sample size 20. Moreover, determining whether were under four 89.1%, 89.72%, 93.08% 91.37% outperforming at show require less across different domains compared methods. performance techniques thoroughly demonstrates feasibility rapidly detecting without need collecting data. research alleviates annotation pressure researchers provides foundation personnel anticipate prevent potential crops.