Plants develop serious of physiological and biochemical responses under stressful environment. Nevertheless, a non-destructive and high throughput method to detect the early crop stress responses is still missing. We hypothesized that using temporal changes in leaf reflectance with machine learning classifier during plant development may allow us to detect early crop stress responses. Here we focus on cauliflower, which was originated from the temperate climate andheat stress can delay or disrupt curd development. By applying heat stress of 40°C for 4 hours, plant height and projected leaf area did not show significant difference between the stressed and control groups after 7 days of heat treatment. However, the curd size in terms of curd diameter of the stressed group is significantly smaller than the control group after 20 days of heat treatment. Two cauliflower cultivars with diverse heat stress responses were used to further study the leaf reflectance between different genetic background. The leaf reflectance were measured at 1, 2, 3, 4, 5, 24 hours after applying heat stress using a portable hyperspectral camera (400 – 1000 nm, 204 spectral bands) under nature light condition. The trend of leaf reflectance and various existing vegetative indices did not show significant differences. On the other hand, using a random forest algorithm by random sampling dataset to repeat the training and prediction procedures, we were able to classify heated and control plants with an overall error rate of ~22%. Model importance analysis showed that the reflectance at wavelength of 397~416 and 697~702 nm are critical to distinguish heated and control plants. These results were not unique to a specific cultivar, highlighting the potential of combining hyperspectral imaging and machine learning to detect early heat stress responses in cauliflower.