Nozzle atomizing performance evaluation in complex environment using residual atrous spatial pyramid network

Conventional spray particle detection methods have disadvantages such as spray field interference, large subjective standard error, and an inability to specifically analyze the spray particle movement. Manual methods used the uniformity of the liquid deposit in the spray chamber to detect spray particles, which only considered the particle density information. Especially, manual detection results by different observers are significantly different, resulting in the low measurement accuracy of the spray particle size. In order to overcome these challenges, this paper proposes a non-contact spray particle segmentation based on the Residual Atrous Spatial Pyramid Network (RASPN). In the RASPN, the spray angle of the fragranced nozzle and the distribution of spray particles of different sizes are evaluated through the statistical method. The experimental results show that the proposed RASPN outperforms the compared methods in terms of detection accuracy. The injection angle is about 31° under an injection pressure of 0.4 MPa, with the highest proportion of 40–80 pixel spray particles.