Detection of spores using polarization image features and BP neural network

Timely detection and control of airborne disease is important to improve productivity. This study proposed a novel approach that utilizes micro polarization image features and a backpropagation neural network (BPNN) to classify and identify airborne disease spores in a greenhouse setting. Firstly, disease spores were collected in the greenhouse, and their surface morphological parameters were analyzed. Subsequently, the micropolarization imaging system for disease spores was established, and the micropolarization images of airborne disease spores from greenhouse crops were collected. Then the micropolarization images of airborne disease spores were processed, and the image features of polarization degree and polarization angle of disease spores were extracted. Finally, a disease spore classification model based on the BPNN was ultimately developed. The results showed that the texture position of the surface of the three disease spores was inconsistent, and the texture also showed an irregular shape. Texture information was present on the longitudinal and transverse axes, with the longitudinal axis exhibiting more uneven texture information. The polarization-degree images of the three disease spores exhibit variations in their representation within the entirety of the beam information. The disease spore polarization angle image exhibited the maximum levels of contrast and entropy when the Gabor filter's direction was set to pi/15. The recognition accuracy of cucumber downy mildew spores, tomato gray mildew spores, and cucumber powdery mildew spores were 75.00%, 83.33%, and 96.67%, respectively. The average recognition accuracy of disease spores was 86.67% based on BPNN and micropolarization image features. This study can provide a novel method for the detection of plant disease spores in the greenhouse.