Enabling smart agriculture by implementing artificial intelligence and embedded sensing

• Published in: Computers & industrial engineering 2022-03, Vol.165, p.107936, Article 107936
• Main Authors: Sharma, Ashutosh, Georgi, Mikhail, Tregubenko, Maxim, Tselykh, Alexey, Tselykh, Alexander
• Format: Article
• Language: English
• Subjects: Artificial neural network; Color normalization; Computational intelligence technique; Genetic algorithm; Varying lighting illuminations
• ISSN: 0360-8352; 1879-0550
• DOI: 10.1016/j.cie.2022.107936

•Image-based analysis using computational intelligence for the prediction of crops.•Crop precision status classification using GA and ANN optimization.•The ANN-based approach is effective to identify wheat crops from unwanted plants.•Proposed method is fast in the prediction and detection of crops with minimum error. The increasing demand of smart agriculture has led to the significant growth and development in the field of crop estimation and prediction improving its productivity. The analysis of crop age status is very important to prevent the excessive fertilization, understand the proper time to harvest and reduce the production cost. Image based analysis using computational intelligence have proved beneficial in estimation of categorical age in the crops. This work focuses on the utilization of predictive computational intelligence technique for the evaluation of nitrogen status in wheat crop. The evaluation depends on the analysis of crop images captured in field at varying lighting illuminations. The wheat crop is initially subjected to HSI color normalization, followed by the optimization process using genetic algorithm (GA) and artificial neural network (ANN) based prediction and crop precision status classification. This ANN based optimized approach can significantly differentiate between the wheat crops from the other unwanted plants and weeds while identifying the crop yield age into categorical classes. The outcomes obtained for the experimentation yields the highest validation accuracy of 97.75% with the minimized error rate of 0.22 and a decrease of 0.28 in the loss value. Comparative to the other contemporary counterparts, the proposed ANN + GA mechanism provides improved performance outcomes while minimizing the error rate.