全國中小學科展

2025年

運用深度學習色彩校正模型之黃疸偵測 Jaundice Detection Using Deep Learning-Based Color Correction Models

現今醫療中,黃疸的早期偵測對肝臟疾病的預防與治療至關重要,但多數人難以在症狀輕微時察覺。我們希望藉由智慧手機影像結合機器學習進行黃疸檢測,提升民眾自我監測的能力。Su 等人(2021)曾使用深度學習和機器學習進行黃疸預測,但其方法依賴專業色卡進行色彩校正,成本高且限制應用範圍。本研究提出以白平衡演算法中的白色補丁法與灰界演算法,搭配深度學習模型 DCCNM1和2 取代色卡,提升黃疸檢測的普及性與便利性。經黃疸偵測效果評估顯示,DCCNM2 在無色卡模型中表現最佳,雖然各指標略低於色卡校正,但其展現出優異的穩定性和準確性,證明其作為無色卡黃疸篩檢方案的可行性。本方法將能提供便捷的居家黃疸檢測途徑,尤其對偏鄉地區居民而言,不僅提升早期發現的機會,還能有效減輕醫護人員的負擔,推動大眾健康管理。

The conspiracy mentality: its relationship with absurdity and ostracism

Conspiracy theories are generally perceived as irrational, absurd and as having a negative effect on our reputation. Yet some people do not see them as such. This may lead us to wonder why and how an individual would come to believe or support such statements. In this work, the primary goal is to obtain the level of absurdity of various conspiracy theories so that they can be used in experiments designed to test Williams' “Strategic Absurdity Hypothesis”. In addition, this study attempts to demonstrate the link between conspiracy mentality and feelings of ostracism. In this research, I analyzed a group of 47 participants recruited via social networks in May 2023. The survey collected their demographic information, their conspiracy mentality, their feeling of ostracism, their knowledge of conspiracy theories and their perception of absurdity toward them. The results were analyzed using correlations and linear regressions. The results show a negative correlation between conspiracy mentality and the perception of absurdity for most theories. In other words, the higher a participant's conspiracy mentality, the less absurd the theory is perceived to be. Therefore, hypothesis (I) is partially supported. On the other hand, hypothesis (II), which says that ostracism predicts conspiracy mentality, is not supported by the results. These results are in line with Williams' “Strategic Absurdity Hypothesis” and Sterelny's signaling theory, explaining that an individual, by expressing agreement with a conspiracy theory typically perceived as absurd, damages their reputation in the eyes of others to show that they belong to the group. However, for the theories relating to COVID-19 and climate change, the conspiracy mentality does not predict the vision of absurdity, suggesting that their currentness and media coverage influence whether they are used as signals. The literature does not fully agree with our results and still presents very divergent opinions regarding the links between feelings of ostracism and conspiracy mentality. Indeed, it remains difficult to determine whether conspiracy mentality influences ostracism or vice versa. In conclusion, this study provides new ideas for future research on the origins and impact of the conspiracy theories.

廣義佩爾方程式的一些探討

這是一份將近持續四年的研究,而這一年佩爾質數的出現,讓我們的討論「突飛猛進」。 佩爾方程式是形如𝑥2−𝑚𝑦2=1的方程式,其中𝑘不為完全平方數之正整數。我們定義廣義佩爾方程式是形如𝑥2−𝑚𝑦2=𝑛 的方程式。在過去的研究中,我們主要從𝑥2−𝑘𝑦2=𝑝 (𝑘,𝑝 皆為互質的奇質數) 的正整數解開始研究,接著延伸到 𝑥2−𝑘𝑦2=2𝑚𝑝1𝑛1𝑝2𝑛2⋯𝑝𝑗𝑛𝑗,進而得到了解的唯一分解性質。而本次的研究,延續之前的工作,對佩爾質數展開了討論。利用蜈蚣彘,我們成功地發現了一些佩爾質數,猜測出一些可能的結果並證明;同時我們對佩爾質數的生成結構做了相當程度的了解。作為結束,設法利用分析的方法解決的之前的問題,以及對方程式的不可約解,是否存在較低次方根解,給出了必要條件。

麥克納姆輪車直線 360 度全向控制與角度誤差之探討

近年來,作者觀察到餐飲業及飯店業引入自主移動機器人(AMR)提供服務,受到廣泛歡迎。不僅如此,隨著AI的風潮,愈來愈多產業引進機器人補足人力的空缺。就輪式機器人而言,麥克納姆輪具相對優異的機動性和運動靈活性,但傳統麥克納姆輪的行進模式只侷限45度倍數的方向。針對此限制,本研究分析傳統麥克納姆輪的合力方向,通過數學推導和實際控制方式的驗證,確認其18個方向移動的數學式,再利用ESP32-S單晶片的WIFI驅動控制系統,實現並驗證 合力分析結果 接著,進一步 建立 麥克納姆輪的全向 360 度直線移動的數學式, 並且用 PWM 脈波調變,實現全向控制。此外,分析地面材質對車子移動的影響,測試出麥克納姆輪適合的地面材質,針對誤差給出可能的校正方式以達更精密的控制。希望就本研究結果擴展未來的應用範圍及使用價值。

Glass Coloring by the production of Colloidal Hydroxide

When doing an experiment to produce colloidal ferric hydroxide, the bottom of the beaker used was colored in yellow-brown with thin film interference. This phenomenon is well-known, but the cause has not been clearly studied. As a result of the research, the coloration on the bottom of the beaker is caused by β-FeOOH forming a thin film which is chemically bonded with Si-OH on the glass surface. Also, the amount of β-FeOOH depends on the number of experiments, the area of the bottom of the beaker, and the concentration of FeCl3 aq. We found that it can be possible to determine the amount of β-FeOOH from the formula m=knsc and the adhesion constant was found to be 6.8✕10-3 (L/m2). In addition, from machine learning we predicted that the thin film thickness becomes thicker as it moves away from the center.

降脂轉肌–將脂肪轉變成肌肉的可能性探討

先前研究發現一種多元不飽和脂肪酸 15-keto-PGE2 能減少肌肉慢性發炎以及增加肌肉生長因子胰島素的敏感性,具有增加肌肉的潛力。我們探討此脂肪酸在小鼠中將脂肪組織轉成肌肉組織的可能性。 我們先以 15-keto-PGE2 治療肥胖型肌少症小鼠後,再以全基因組mRNA定序,發現在小鼠皮下脂肪中,肌肉特異基因表現量高度上升。透過基因表現路徑分析軟體 GSEA證實此脂肪酸會高度引發皮下脂肪中,與肌肉分化、肌肉收縮與肌肉結構相關的生物路徑。另外, 15-keto-PGE2 也可以誘導脂肪前驅細胞3T3-L1中與肌肉生成相關的基因表現。 最後,我們發現以 15-keto-PGE2 治療的老年肥胖型肌少症小鼠,體重、血糖、脂肪比例下降,肌肉質量及力量上升,證實了它在體內同時減少脂肪並增加肌肉量的效果。

自監督學習在臺灣手語辨識上之應用研究

在臺灣手語辨識,先前研究所使用的監督式學習需要大量標記樣本而限制可辨識詞彙量。為此,本研究借鑒自然語言處理領域中BERT 的遮罩想法,將未標記手語影片隨機遮蓋部分幀數,並讓模型學習預測被遮蓋的幀數以學習臺灣手語的特徵,並透過遷移學習來訓練辨識模型,此作法可克服現有臺灣手語資料缺少的問題。經過實驗,本研究訓練之詞彙辨識模型達成了242 個詞彙量,94.8%的準確率。 此外,先前研究皆未在手語句子翻譯上有成果。因此本研究基於預訓練模型,整合設計手語翻譯的系統,實驗中,系統在100 個句子的翻譯表現達到88%的準,且BLEU-4 分數取得20.98,證明自監督學習的方式在手語辨識、翻譯上是有效的。並展現出樣本需求少與辨識詞彙量可輕易擴大的潛力。

Wetting Tracing Paper—Fiber Porous Media Curling Behavior and Mechanisms

This research presents a novel approach to understanding the curling and uncurling behavior of tracing paper when exposed to water, identifying limitations in traditional diffusion-based models like Fick’s second law. While Fick's model adequately represents the uncurling phase, where water content is stable, it falls short during the curling phase due to its inability to account for dynamic changes in diffusivity. Our study identifies capillary action, modeled through Richards' equation, as the primary mechanism in the curling phase, where diffusivity varies with water content due to capillary-driven water movement through the paper's porous structure. Experimental data align well with the Richards' equation model, highlighting a saturation point where curvature peaks, governed by evaporation's impact on moisture balance. To simulate this phenomenon, we developed a finite difference approximation scheme based on Richards' equation, discretizing the spatial domain for detailed control over moisture dynamics and incorporating the Robin boundary condition with virtual points. This approach, combined with evaporation considerations, produces simulation results consistent with observed data, emphasizing evaporation’s role in steady-state moisture gradients and the subsequent deformation mechanics. Our findings further reveal that factors like paper thickness, temperature, and salt concentration significantly influence curling behavior. We established linear correlations between peak time and thickness reciprocal, as well as between peak curvature and thickness squared, supporting theoretical models. Temperature affects both peak curvature and curling rate due to changes in viscosity and surface tension, and higher temperatures prevent full uncurling due to sustained evaporation effects. Increased salt concentration heightens peak curvature without altering expansion ratio, suggesting additional variables in play.

基於心電圖的智慧睡眠分析

睡眠相關問題常見於現代緊張的社會,傳統睡眠分析方法需要腦電圖(EEG)、肌電圖(EMG)、眼電圖(EOG)等信號,量測複雜度高。本研究透過 Python 程式語言以深度學習和階層式投票機器學習方法,開發一套自動分析程式,僅透過心電圖(ECG)信號分析睡眠階段。並結合睡眠評估標準,製訂一可量化的睡眠品質評估表,提供臨床醫師判讀睡眠品質的指標。本研究的優點是僅透過一種信號便能準確、客觀、快速分析,且操作介面簡易。研究結果顯示,本研究清醒和睡眠狀態之辨識準確率高達約90%,與其他類似睡眠品質評估研究的論文比較,準確率高出10~17%,整體睡眠階段分析準確度高達87%。本研究方法未來可應用於臨床醫療,協助醫師做精準的患者睡眠品質診斷。

法拉第波輔助合成奈米鎳並應用於有機污染物的快速脫色 Nanostructured Nickel Synthesized through Faraday Waves and Its Application to Rapid Contaminants Decolorization

超音波已廣泛用於奈米粒子的製備,然可聽聞音對奈米粒子製備的影響卻少有研究。本研究以簡易喇叭裝置產生可聽聞音並在溶液表面產生法拉第波及內部流動,來輔助製備奈米鎳。法拉第波是一種表面非線性駐波,透過調整容器形狀、振動頻率等,可產生不同波形。本研究嘗試在法拉第波輔助下,以化學還原法及電沉積法製備出不同性質的奈米粒子。SEM量測並比較無輔助、法拉第波輔助、超音波輔助製備出的奈米鎳的形貌、分布的差異。並將其應用於有機物(即剛果紅、亞甲藍、4-硝基苯酚、2-硝基苯酚)之催化還原。而由SEM量測、催化還原結果及理論模擬反應熱可知,法拉第波確實能夠改善奈米鎳的粒徑大小、分散性、對氫的吸附能力及催化還原能力。