Police Evidence Analysis and AI. Exploring practical developments and perspectives on artificial intelligence. https://kimkj.com/podcast/police-evidence-ai-en en Kyungjin Kim Kyungjin Kimkimkj008@gmail.com episodic false A compilation for 31 October 2026 built from material provided by attorney Kyungjin Kim, focused on introducing and explaining the original sources. 1. Lausanne and the Belgian Royal Military Academy validate hierarchical machine learning on portable NIR scans for on-site explosive identification 2. Ankara University validates YOLOv12-based automated fracture detection for forensic anthropology identification 3. Inks, sedatives and cannabis: three forensic chemistry studies combining spectroscopy with machine learning 4. Urock demonstrates the DFAS multimodal AI digital forensics suite at RISCON Tokyo 2026 Each segment was generated separately as a Brief from a single selected source and then joined in order. The programme introduction appears only in the first segment. Claims, figures and assessments belong to the cited sources, not to attorney Kim. Audio was AI-generated. The closing provider notice at the end is a Korean AI voice line. kimkj-police-evidence-ai-en-20261031 Sat, 31 Oct 2026 09:38:16 +0900 323 false An English compilation of four individually selected NotebookLM Briefs based on material provided by attorney Kyungjin Kim. 1. Switzerland: ICTjournal, 7 October 2026, on Deeplink's Trinity at the Morges regional police in Vaud. About 50 officers have used it mandatorily since spring 2026. Dispatch records, voice notes and scene photos become draft logs/reports and suggested legal provisions. The source describes an isolated network built for data protection. Officers review, edit and sign, retain legal responsibility, and the reports may be used in court. The police-reported figures are a 40–50% administrative workload, an initial goal of a 10% EFFICIENCY GAIN, and logs formerly taking 30–45 minutes now taking LESS THAN HALF that time. 2. China: People's Daily, 29 October 2026, on Jinjiang in Fujian and its Zhaqianshao anti-telecom-fraud system. Multi-model machine learning and AI agents support risk identification, customised warning scripts and follow-up. The June 2026 customer-service impersonation case combines AI assistance with a police/bank payment freeze and a reported CNY 1.22 million protected. Agency-reported figures are 9,261 CASE samples, 84.12% PRECISION, prevention calls averaging 6 to 30 seconds and outreach to 30,000 medium/high-risk citizens. The September 9 Ministry of Public Security competition result was second place. Model retraining and testing generalisation remain the source's own limits. 3. Forensics: Bioengineer.org, 28 October 2026, reports early research by Hussain and colleagues in the Egyptian Journal of Forensic Sciences. The sample was 240 living adults, 120 men and 120 women, aged 18–40. ResNet50/VGG16/EfficientNet-B0/DenseNet121 scored 77/73/69/62 percent. One population, self-reported sex and untested fire/decomposition effects limit this sex-estimation study; it is not identification. Separately, Boan News, 27 July 2025, describes the past Korean ATRIP case: AI first screens short-circuit-mark photos and an NFS examiner gives final review. The source reports appraisal turnaround from 3–4 weeks to within hours in that 2025 case. 4. South Africa: Biometric Update, 15 September 2026, on SITA's three-year SAPS body/dash camera supply tender RFB 3286-2026. Procurement is not deployment. Device-access biometrics and video analytics are distinct; ANPR, motion detection and platform alerts are specifications. Quantity and budget are unstated. The source's POPIA police-processing exceptions, disputed scope of NPR linkage, civil-society concerns and Information Regulator warnings are part of the reported policy discussion. Operational effects remain unconfirmed and legal/data-linkage scope is under discussion. AI-generated and edited; four distinct subtopics, one selected source per Brief, only the first show intro/material credit. Source claims and historical/planned/research stages are distinguished. No audio-speed changes or silence added for length. Existing GUID and original publication date retained. The new-outro policy is not retrofitted to this existing r2 revision. kimkj-police-evidence-ai-en-20261029-trial Thu, 29 Oct 2026 14:28:04 +0000 311 false