Maize seeds forecasting with hybrid directional and bi-directional long short-term memory models

Area of Science:

• Agricultural Science
• Computer Science
• Biotechnology

Background:

• Seed purity is a critical factor influencing crop yield.
• Accurate classification of maize cultivars is essential for maintaining seed purity and maximizing agricultural output.
• Current methods for maize cultivar identification face challenges in accuracy and efficiency.

Purpose of the Study:

• To develop and evaluate advanced machine learning models for accurate maize cultivar classification.
• To investigate the effectiveness of deep learning architectures, specifically AlexNet and ResNet50, combined with recurrent neural networks for image-based classification.
• To identify the optimal model for high-success classification of four distinct maize cultivars.

Main Methods:

• A specialized dataset comprising 14,469 images across four maize cultivars (BT6470, CALIPOS, ES_ARMANDI, HIVA) was curated.
• Transfer learning using pre-trained AlexNet and ResNet50 architectures was employed for image feature extraction.
• Hybrid models were created by integrating AlexNet and ResNet50 with Long Short-Term Memory (LSTM) and Bi-directional Long Short-Term Memory (BiLSTM) algorithms.

Main Results:

• Six distinct classification models were designed and evaluated.
• The ResNet50 architecture, when hybridized with the Bi-directional Long Short-Term Memory (BiLSTM) algorithm, demonstrated superior performance.
• The ResNet50+BiLSTM model achieved the highest classification accuracy, reaching 98.10%.

Conclusions:

• Hybrid deep learning models integrating convolutional neural networks and recurrent neural networks offer significant improvements in maize cultivar classification.
• The ResNet50+BiLSTM model presents a highly effective solution for automated and accurate identification of maize varieties.
• This approach has the potential to enhance seed quality control and contribute to increased agricultural productivity.