Can We Believe the Rumors Circulating about the Asian Giant Hornet?
我们能相信流传的关于亚洲大黄蜂的谣言吗？

Abstract  抽象

Summary The Asian giant hornet, an invasive pest, was discovered on Vancouver Island in British Columbia, Canada. Subsequently, many residents claimed to have witnessed the hornets, but the authenticity of these claims remains to be verified. To better assess the spread of the Asian giant hornet, we modeled ResNet to identify images and predict the likelihood of misclassification.
摘要 亚洲大黄蜂是一种入侵性害虫，是在加拿大不列颠哥伦比亚省的温哥华岛发现的。随后，许多居民声称目睹了大黄蜂，但这些说法的真实性仍有待核实。为了更好地评估亚洲大黄蜂的传播，我们对 ResNet 进行了建模以识别图像并预测错误分类的可能性。

For task(a), we modeled prediction system with Prophet and SARIMA based on the collected public reports, to predict the time - dependent spread characteristics of the Asian giant hornet. The MSLE of Model 1 reached 2.89, and the MAE of Model 2 reached 9.00, indicating good predictive performance of the models. Both models suggest that the number of Asian giant hornet nests exhibits a spiky peak - shaped trend over time. In August 2020, the number of Asian giant hornet nests reached its maximum value, during which their activity level was high. Their active period is from May to October, with a cycle of one year.
对于任务 （a），我们根据收集到的公开报告，使用 Prophet 和 SARIMA 对预测系统进行建模，以预测亚洲大黄蜂的时间依赖性传播特征。模型 1 的 MSLE 达到 2.89，模型 2 的 MAE 达到 9.00，表明模型具有良好的预测性能。这两个模型都表明，随着时间的推移，亚洲巨型大黄蜂巢的数量呈现出尖刺的峰值形状趋势。2020 年 8 月，亚洲大黄蜂巢的数量达到最大值，在此期间它们的活动水平很高。他们的活跃期是从 5 月到 10 月，以一年为周期。

For task(b), we utilized the pre - trained ResNet model and adjusted relevant parameters to identify 3,305 images provided by the public. By introducing functions such as the classification function, cross - entropy, and Adam optimization algorithm, we successfully trained the model, achieving an accuracy of $95.86\mathrm{\%}$$95.86\mathrm{\%}$95.86%95.86 \%. This model can accurately distinguish between wasps and hornets based on images and evaluate the likelihood of misclassification.
对于任务 （b），我们利用预先训练的 ResNet 模型并调整相关参数，以识别公众提供的 3,305 张图像。通过引入分类函数、交叉熵和 Adam 优化算法等函数，我们成功地训练了模型，实现了 的准确率。 $95.86\mathrm{\%}$$95.86\mathrm{\%}$95.86%95.86 \% 该模型可以根据图像准确区分黄蜂和大黄蜂，并评估错误分类的可能性。

For task©, based on the models from Task 1 and Task 2, we performed a weighting process, assigning weights of $30\mathrm{\%}$$30\mathrm{\%}$30%30 \% and $70\mathrm{\%}$$70\mathrm{\%}$70%70 \% respectively. The newly generated model is used for classification to select the reports that are most likely to be positive sightings.
对于任务©，根据任务 1 和任务 2 中的模型，我们执行了加权过程，分别分配了 $30\mathrm{\%}$$30\mathrm{\%}$30%30 \% 和 的 $70\mathrm{\%}$$70\mathrm{\%}$70%70 \% 权重。新生成的模型用于分类，以选择最有可能是正面目击事件的报告。

For task(d) and task(e), when additional new reports are added, we can manipulate the ResNet to train a model with a larger dataset and increase the update frequency. When the evidence provided in a certain month shows that no correct cases of the Asian giant hornet have been observed, it is determined that this pest has been eradicated in Washington State.
对于 task（d） 和 task（e），当添加额外的新报告时，我们可以操纵 ResNet 来训练具有更大数据集的模型并提高更新频率。当某个月提供的证据显示没有观察到亚洲大黄蜂的正确案例时，可以确定这种害虫已在华盛顿州被根除。