Have library access?
IP:3.236.143.154
  • Journals

AI辨識畫作即時反映情緒提升青少年輔導效能

AI Recognizes Paintings Reflecting Emotions Improves Emotional Counseling for Teenagers

Abstracts


針對青少年情緒問題之即時性處理,需要輔導諮商人員的持續性關注與投入,傳統文字類量測工具容易造成受測者規避真實狀態,故越來越多心輔專業人士運用房樹人繪畫測驗瞭解受測者之心理狀況,但判讀畫作時間較長,且當數量多時,較容易錯失輔導黃金時刻,故本研究以房樹人繪畫測驗作為心理量測工具,輔以人工智慧協助輔導諮商人員兼具科學依據與效率進行判讀畫作,並即時幫助情緒失序的青少年,以防憾事發生。本系統之開發流程分為三階段,第一階段透過單因子變異數分析(ANOVA)與多重比較(Post Hoc)驗證文字型與繪畫型心理測驗之差異及本研究之房樹人繪畫測驗AI辨識系統與房樹人繪畫測驗專家學者之結果比對調查系統開發可行性,第二階段以YOLOv4模型、Python Flask及OpenCV DNN模塊執行系統訓練實作,第三階段以YOLOv7模型並加入OpenVINO套件執行系統優化。研究結果發現系統開發第一階段繪畫型心理測驗相較文字型較能得知受測者真實心理,第二階段採房樹人繪畫測驗專家學者提供之數據,訓練後模型全類平均準確率(mean average precision, mAP)為57.66%,而第三階段之全類平均準確率(mAP)則為64.55%。

Parallel abstracts


Treating adolescent emotional issues promptly necessitates counselors' sustained attention. Conventional text-based measures risk evading true feelings. Hence, experts increasingly employ the House-Tree-Person (HTP) drawing test to grasp subjects' psychological state. However, interpreting HTP drawings is time-consuming, risking missed counseling opportunities with numerous drawings. This study employs the HTP test as a psychological measure, coupled with AI support for efficient analysis. This aids counselors in providing timely help to emotionally troubled youths, averting potential regrets. The development process of this system consists of three stages. In the first stage, the differences between text-based and drawing-based psychological tests are verified, and the results of the AI recognition system for the HTP test in this study are compared with those of expert scholars in the field. The second stage involves the implementation of system training using YOLOv4, Python Flask, and OpenCV DNN. In the third stage, YOLOv7 and OpenVINO are integrated for system optimization. The research findings indicate that in the first stage of system development, drawing-based psychological tests provide a better understanding of the true psychological state of the test subjects compared to text-based tests. In the second stage, using data provided by expert scholars in the HTP test, the trained model achieves an overall mean average precision (mAP) of 57.66%. In the third stage, the overall mAP further improves to 64.55%.

References


Alex, K., Ilya, S., Geoffrey, E. H. (2012). ImageNet classification with deep convolutional neural networks. Adv Neural Inf Process Syst,25,1106–14. doi:10.1145/3065386.
Burns, B.J., Costello, E.J., Angold, A., Tweed, D., Stangl, D., Farmer, E.M., Erkanli, A. (1995). Children’s mental health service use across service sectors. Health Affairs (Millwood),14(3),147-159. doi: 10.1377/hlthaff.14.3.147.
Ezra, G., Hefei, W., Benjamin, F. M. (2020). Coronavirus Disease 2019 (COVID-19) and Mental Health for Children and Adolescents. JAMA Pediatrics, 174(9),819-820. doi:10.1001/jamapediatrics.2020.1456.
Nunnally, J. C. (1978). Psychometric theory (2nd Ed.). New York: McGraw-Hill.
一沙(2016)。OPEN MIND ! 房樹人繪畫心理學測出你內心深處的秘密與焦慮。臺北市:橡樹林文化。

Read-around