首页|基于视觉大模型技术的数智创新研究——以西湖景区南方竹林的生态警务为例

基于视觉大模型技术的数智创新研究——以西湖景区南方竹林的生态警务为例

扫码查看
[目的]以西湖景区南方竹林的生态警务为对象研发"数智创新生态警务系统",包括竹林生态保护、野生动物保护、旅游景点服务、灾害预测预防、人员救援处置等功能.[方法]基于视觉大模型技术,选取杭州西湖风景名胜区五云山竹产区为实践点,使用当下最先进的机器自监督学习技术进行研究.具体做法为通过预训练大模型,包括多个基础视图数据微模型,在微模型基础上叠加客观世界逻辑,结合竹林生态警务场景最小化样本数据,形成各类实战的视觉算法模型.[结果]突破了传统单一化视觉算法无法适应多场景、样本需求高、训练时间长等制约条件,促成视频全解析成为行业标配,快速生成匹配算法并落地各类应用场景,如竹林违规砍伐识别算法、大型野生动物识别算法、珍稀动物识别算法、人员轨迹聚类算法、异常侵入识别算法等竹林景区特殊场景模型.[结论]改变了当前科技警务机制下对视觉算法模型需求的大机制;不再依赖于各行业算法科技公司耗时耗力训练算法的模式,每个警务人员都可以是算法的训练者,并结合自身需求,训练符合工作业务的算法模型;推动了"技术、机制、体制"三轮驱动,将理论技术赋予实战,达成了"科技生态警务"的目标.
Research on Digital Intelligence Innovation Based on Visual Large Model Technology,A Case Study of Ecological Policing in the Southern Bamboo Forests of West Lake Scenic Area
[Objective]The objective of this study is to develop a "digital intelligence innovative ecological policing system"in the bamboo forests in the southern part of the West Lake Scenic Area,which includes the functions of ecological protection of the bamboo forests,wildlife protection,tourist attraction services,disaster prediction and prevention,and personnel rescue and disposal.[Method]Based on the visual large model technology,the Wuyun Mountain production area in the West Lake Scenic Area of Hangzhou was selected as the practice point,and the most advanced machine self-supervised learning technology was used for research.The specific method was to pre-train the large model,including a number of basic view data micro models,superimpose the objective world logic on the basis of the micro model,and combine the minimized sample data of the bamboo forest ecological policing scene to form various practical visual algorithm models.[Result]The results broke through the constraints of the traditional single vision algorithm that could not adapt to multiple scenes,high sampledemand,and long training time etc.,and promoted the full video analysis to become the industry standard,and quickly generated matching algorithms which were implemented in various application scenarios,such as recognition algorithm of illegal logging in bamboo forests algorithm of large wildlife identification,recognition algorithm of rare animals,algorithm of human trajectory clustering,algorithm of abnormal intrusion recognition and other special scene models of scenic spots in bamboo forests.[Conclusion]The large mechanism of the demand for visual algorithm model under the current scientific and technological policing mechanism has been changed.And there's no need to rely on the time-consuming and labor-intensive training mode of algorithm technology companies in various industries,each police officer can be a trainer of algorithms,and combine their own needs to train algorithm models that meet the needs of their work.It has promoted the three-wheel drive of"technology,mechanism and system",given theoretical technology to practice,and achieved the goal of"science and technology ecological policing".

Bamboo forest ecologyEcological policingVisual big modelSelf supervisedlearningPractical innovation

田伟

展开 >

杭州市公安局 西湖风景名胜区分局,浙江 杭州310007

竹林生态 生态警务 视觉大模型 自监督学习 实践创新

2024

竹子学报
国家林业局竹子研究开发中心,中国林学会竹子分会,浙江省林业科学研究所

竹子学报

CSTPCD
影响因子:0.535
ISSN:1000-6567
年,卷(期):2024.43(3)