Fu xiaoyu ^1* Han Wu ^2* ZhongJie Shen ^3*

• ¹ Zhongnan University of Economics and Law, Wuhan, China
• ² Wuhan Vocational and Technical College, Wuhan, China
• ³ Xijing University, Xi'an, Shaanxi Province, China

The chemical manufacturing industry is developing rapidly, and its manufacturing and transportation pose high environmental risks. Therefore, environmental risk assessment methods have been studied. The sensitivity matrix algorithm has a high degree of specificity for environmental risk assessment data, so it ignores the carrying properties of chemicals, which can easily cause harm to researchers in experiments. To address this issue, this study introduces toxicity feature distance into the sensitivity matrix algorithm for feature analysis of chemical exposure toxicology and applies it to risk assessment in multi-media environments to generate a fusion model. To verify the performance of the model, this study conducts experiments on the Toxic dataset and compares it with the experimental results of three models, including convolutional neural networks, to verify the superiority of the fusion model. The experiment found that in multi-media environment risk assessment, the F1 value of the research method reached stability at 120 seconds, reaching 7.4.When conducting system accuracy analysis, the research method achieved the highest system accuracy, reaching 94.7%. In the experiment of chemical substance radioactivity, the research method showed a chemical substance radioactivity of 46mGs at a system participation degree of 60. The experimental data show that the proposed fusion model has superior performance in chemical exposure toxicology and can effectively face risk assessment in multimedia environments.