国内外7款常用植物识别软件应用效果评价
摘要:
植物识别软件在植物资源专业性调查、科普及教学等方面的应用越来越受到重视，同时也成为人们日常生活中可灵活便捷识别周边植物的有力工具。根据植物食用性、药用性、观赏性等特点，通过实地拍摄结合中国植物志图像资源，筛选含有多数科、属中代表性植物163科379属618种植物，建立测试图像数据库，选择形色、花伴侣、花伴侣专业版、PlantNet、LeafSnap等国内外7款常用植物识别软件进行应用效果分析评价。研究结果表明：植物识别得分最高的为花伴侣专业版，最低的为LeafSnap，同时也发现各软件植物识别效果之间差异并不显著；从植物不同器官等方面识别来看花伴侣专业版表现最好，LeafSnap识别准确率最低；7款软件对被子植物的识别准确率高于裸子植物和蕨类植物，对栽培植物的识别准确率高于对野生植物的；植物识别结果输出稳定性方面PlantNet最佳，形色次之。综合来看，形色作为识别花卉果蔬的常用植物识别软件可用于日常生活和常规的植物教学及科普等方面；其在野外专业性调查等方面结合花伴侣专业版或专家识别可提升识别效果。随着人工智能发展，各款植物识别软件的特色和识别能力还有很多提升空间，将有利于促进植物资源调查与保护、植物教学与科普教育等方面的发展。
Abstract:
The application of plant identification softwares in professional investigation, science popularization, and teaching of plant resources is increasingly valued, and it has also become a powerful tool for people to flexibly and conveniently identify the surrounding plants in their daily lives. Based on the characteristics of plant edibility, medicinal use and ornamental value, the representative plants of 163 families, 379 genera, and 618 species were selected through field photography combined with Chinese plant image database resources, and a test image database was established. Seven commonly used plant identification softwares, including PictureThis, FlowerMate, FlowerMate2.0, PlantNet, LeafSnap, etc. at home and abroad, were selected for application effect evaluation. The results showed that FlowerMate2.0 had the highest identification score and LeafSnap had the lowest identification score. However, no significant difference was found among the selected software for plant identification accuracy. From the aspect of different plant organs, the identification of FlowerMate2.0 showed the best performance, while LeafSnap had the lowest performance. The identification accuracy of six softwares for angiosperms was higher than that of gymnosperms and ferns, and the identification accuracy for cultivated plants was higher than that for wild plants. PlantNet had the best output stability in plant identification, followed by PictureThis. Overall, PictureThis as the commonly used plant identification software for identifying flowers, fruits, and vegetables, shape and color can be used in daily life and routine plant teaching and science popularization. Combining professional field surveys with FlowerMate2.0 or expert identification can improve identification accuracy. With the development of artificial intelligence, the characteristics and identification effects of various plant identification softwares will be further improved, which will be conducive to promoting the development of plant resource investigation and protection, plant teaching and popular science education.