Revolutionizing Agriculture with AI: Crop Planting Density Optimization System

By Dr. Sampath LonkaReviewed by Susha Cheriyedath, M.Sc.Sep 27 2023

In a recent publication in the journal Agronomy, researchers introduced the Crop Planting Density Optimization System (CPDOS), an intelligent online system employing artificial intelligence (AI) techniques such as genetic algorithms and neural networks.

Background

Crop planting density, influencing population, light utilization, and yield profoundly impacts agriculture. Research has long explored density-yield relationships. Planting density and fertilizer application significantly affect crop yield and quality. Studies demonstrate their impact on attributes such as photosynthetic capacity and grain quality. Agriculture leverages technology, such as crop monitoring systems. González-Esquiva's system uses image analysis for irrigation decisions. Challenges persist due to the nonlinear nature of yield-density models, limiting their widespread use.

Crop Planting Density Optimization System

The current study employs modern computing to optimize planting density, enhancing crop planning. Genetic algorithms optimize classical yield-density models. An economic-yield-density relationship is established, leading to the development of a crop planting density optimization system (CPDOS) with three core modules.

The CPDOS utilizes a genetic algorithm, dichotomy, polynomial regression, and a neural network. Six classical yield-density models were obtained, and the genetic algorithm determined the best-fit model, accounting for crop production costs. The optimal planting density range was calculated using dichotomy. A neural network and polynomial regression were used to analyze planting density and fertilization data. A genetic algorithm optimized the combination of planting density and fertilizer for maximum yield.

Implementation of CPDOS: CPDOS adopts a browser and Server system structure with three layers: the system front end, the system back end, and the system database. CPDOS provides user-friendly interfaces and data visualization capabilities, encompassing user login, yield-density model optimization, optimal planting density range calculation, and fertilization or planting density optimization. CPDOS enables users to obtain yield-density equations, optimal yield-density equations, and planting density and fertilization ratios. Data visualization includes charts, model curves, and goodness-of-fit diagrams.

Crop Yield-Density Models: Various yield-density models are employed, including progressive, parabolic, and mixed models. Different parameters are used in these models to describe the relationships between planting density and crop yield. Quadratic expressions and linear models are also considered.

CPDOS Parameter Estimation of Yield-Density Model: Researchers explored CPDOS Parameter Estimation of Yield-Density Models and the Selection of Optimal Yield-Density Models using an Evolutionary Genetic Algorithm (EGA). Six nonlinear yield-density models are considered; the challenge lies in parameter estimation. Different nonlinear models often require different parameter estimation methods. Researchers aim to find a parameter estimation method that is not dependent on the specific nonlinear model expression, making it suitable for various nonlinear systems. They utilize EGA combined with the least-squares method to estimate parameters for the six chosen yield-density models. They evaluate the models using the mean square error (MSE) as the selection criterion to identify the best-fitting model for the given data. This approach's effectiveness is verified through simulation results compared to other methods.

CPDOS Genetic Algorithm EGA Strategy Design: It includes using an elite retention strategy to optimize the classical genetic algorithm, improving its global convergence and robustness. The paper also delves into coding and population initialization, setting the fitness function based on the least-squares method, selection strategies using tournament selection, crossover strategies with double-point crossover, mutation strategies, and the elite retention strategy to enhance the search process.

Additionally, the study uses Geatpy, an evolutionary algorithm toolbox, for parameter estimation and model screening. The neural network is applied to determine the most suitable planting density range for the highest economic benefit of crops. Cross-validation is used to handle limited experimental data.

Results and Analysis

The study employed a genetic algorithm with an elite retention strategy for parameter estimation. Utilizing binary coding, a population size of 200, and the tournament operator for individual selection, a two-point crossover with a 0.7 probability and a basic bit mutation at a 0.2 probability were performed. Iterations were capped at 10,000.

The researchers employed the progressive yield-density model to fit the yield-density data. Comparisons between the results of this model fitting and Mo Huidong's work were conducted using the EGA and sample estimation methods. EGA significantly improved accuracy, reducing MSE by 24.77 percent compared to Mo Huidong's fitting and increasing the coefficient of determination (R square) by 1.85 percent. This highlights the efficacy of EGA for parameter estimation.

Considering crop planting costs, an economic crop yield equation was derived using a genetic algorithm. EGA-based model fitting outperformed Mo Huidong's original method, reducing MSE by 61.019 percent and increasing R square by 5.0501 percent. The economic yield-density model was established, and the optimal planting density was determined through dichotomy. The optimal planting density range was calculated as [2.118476, 3.869274] based on economic criteria.

For the CPDOS, a neural network yielded high accuracy in fitting planting density, fertilization, and yield relationships, with a root mean square error (RMSE) of 29.2586 and an R square of 0.9790. This approach can be applied to various crops, providing optimal yield-density models and insights into the planting density range.

Conclusion

In summary, researchers proposed the CPDOS system for optimal planting density and research outcomes. It relies on the EGA, dichotomy, polynomial regression, and a BP neural network. The CPDOS features three core modules: yield-density optimization, optimal planting density range, and fertilization and planting density optimization. It aids users in analyzing yield density and related data, optimizing planting density, and fertilizer allocation.