Determination of aquaponic water macronutrient concentrations based on lactuca sativa leaf photosynthetic signatures using hybrid gravitational search and recurrent neural network

Crop quality depends dominantly on the nutrients present in its growth media. For precision farming, fertigation is a challenge, especially when dealing with economical and efficiency factors. In this study, the aquaponic pond water macronutrient prediction model (wNPK) was developed based on leaf photosynthetic signature predictors. Aquaphotomics was preliminarily used for correlating physical limnological properties with nitrate, phosphate, potassium concentrations, and the leaf signatures. Using a digital camera, 18 spectro-textural-morphological features were extracted. Neighborhood component analysis (NCA) and ReliefF algorithms selected the spectral components blue, a*, and red minus luma as the most significant as supported by principal component analysis, resulting in low computational cost. A Gravitational Search Algorithm (GSA) was employed to optimize the recurrent neural network (RNN) architecture resulting in higher sensitivity. The hybrid NCA-ReliefF-GSA-RNN (wNPK) predicted NPK with 93.61, 84.03, and 91.39 % accuracy, respectively, besting out other configured feature-based machine learning models. Using wNPK, it was confirmed that potassium helped in accelerating seed germination and nitrogen in promoting chlorophyll intensification, especially on the 6^th week after sowing. Phosphate and potassium were the energy and health elements that were consumed in a larger amount at the end of the head development stage. wNPK rules out that macronutrient concentration have a direct resemblance to crop leaf signatures; thus, a leaf is a good indicator of the water quality. The results pointed out that the use of a single camera to measure both water macronutrient concentrations and crop signature at the same time is an innovative, efficient, and economical approach for precision farming.