From Event: SPIE Defense + Commercial Sensing, 2023
Escherichia coli (E. coli) bacteria when transmitted to humans by infected fruits like citrus can cause serious illnesses like bloody diarrhea and kidney disease (Hemolytic Uremic Syndrome). Therefore, inspection for the presence of E.coli colonies on fruits and vegetables would be beneficial for food safety if an appropriate sensor and detection approach were available. In this paper, we evaluate the efficacy of SafetySpect’s Contamination, Sanitization Inspection, and Disinfection (CSI-D+) system, which consists of an UV camera, an RGB camera, and illumination at two fluorescence excitation wavelengths, ultraviolet C (UVC) at 275 nm and violet at 405 nm, respectively. For this study, bacterial population at four different concentration levels were inoculated on black rubber slides. The black rubber slides provided a fluorescence-free background for benchmark tests on E.coli-containing droplets, since even fruit peels have a subdued fluorescence response which appears as low intensity noise in the background of the UV images. To increase the number of image datasets to avoid overfitting issue of deep leaning and machine learning models, synthetic datasets were generated using StyleGAN2- ADA generative network. As a first approach, the fluorescence images were denoised using discrete wavelet transform (DWT) and reconstructed denoised images were used to train StyleGAN2-ADA for generating larger datasets which were later used by VGG19 with SoftMax classifier. In the second approach, VGG19 was used to extract features and then Linear and RBF SVM were used for classification. It was found that using Symlets family of DWT denoised images at an average PSNR value of 25.59 dB. VGG19 with SoftMax was able to classify the images at overall accuracy of 84% without synthetic dataset and at 94% by using augmented dataset generated by StyleGAN2-ADA. It was also found that using RBF SVM increased the accuracy by 2% to 96% and linear SVM by 3% to 97%. The findings reported here could be used to detect presence of E. coli bacterial population on citrus peels for taking the necessary actions for decontamination.
© (2023) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Pappu Kumar Yadav, Thomas Burks, Quentin Frederick, Jianwei Qin, Moon Kim, Mark A. Ritenour, and Kunal Dudhe, "Classifying E. coli concentration levels on multispectral fluorescence images with discrete wavelet transform, deep learning, and support vector machine," Proc. SPIE 12545, Sensing for Agriculture and Food Quality and Safety XV, 1254508 (Presented at SPIE Defense + Commercial Sensing: May 02, 2023; Published: 13 June 2023); https://doi.org/10.1117/12.2663933.