Using convolutional neural networks to classify the maturity of compost based on sewage sludge and rapeseed straw

Abstract

Large amounts of municipal sewage sludge are generated every year. Composting is one of the most popular and cost-effective methods of handling this waste. It is particularly important to carry out the composting of sewage sludge in an efficient way. The identification of the stage of early maturity of composted material is crucial. The objective of this study was to check the possibility of using convolutional neural networks to classify the early maturity of compost made from sewage sludge and rapeseed straw. The literature review showed that this specific and modern type of neural network had not been investigated for this purpose before. A mixture of these two substrates, composted for 25 days in 7 experiments, was the research material. Images of the material samples were acquired in a dedicated photo station illuminated with: visible light, ultraviolet light, and mixed visible and ultraviolet light. For each acquisition variant 1312 images were obtained. On the basis of the acquired images, 25 convolutional neural networks with one convolutional layer and a varied number of convolution filters were developed to determine the stage of early maturity of the material. Their classification error ranged from 0.51% to 17.77%. The best classification result was obtained for the model containing 16 convolution filters and based on images of the material acquired in mixed visible and ultraviolet light. The results presented in this paper have shown that convolutional neural networks are an adequate tool to assess compost maturity. Such models may provide a simplified and cost-effective approach to the approximate evaluation of the early maturity of compost.