Plant Disease Detection Strategy Based on Image Texture and Bayesian Optimization with Small Neural Networks

Abstract

A novel method of disease diagnosis, based on images that capture every part of a diseased plant, such as the leaf, the fruit, the root, etc., is presented in this paper. As is well known, the plant genotypic and phenotypic characteristics can significantly impact how plants are affected by viruses, bacteria, or fungi that cause disease. Assume that these data are unknown at the outset and that the appropriate precautions are not taken to prevent classifications skewed toward uninteresting traits. An approach to avoid categorization bias brought on by the morphology of leaves is suggested in this study. The basis of this approach is the extraction of textural features. Additionally, Bayesian Optimization is suggested to obtain training hyperparameters that enable the creation of better-trained artificial neural networks. First, we initially pre-processed the images from the PlantVillage dataset to remove background noise. Then, tiles from images were used to reduce any potential bias from leaf form. Finally, several cutting-edge tiny convolutional neural networks (CNNs), created for contexts with little processing power, were trained on a new dataset of 85 × 85 × 3 px images. MobileNet, which had a 96.31% accuracy rate, and SqueezeNet, which had a 95.05% accuracy rate, were the models that predicted the best performance. The results were then examined using Precision and Recall measures, which are important for identifying plant diseases.

A novel CNN gap layer for growth prediction of palm tree plantlings

“…The GAP layers [51] output is projected into a classifier called SoftMax. Bayes optimization tests [52] also confirmed that ReLU and dropout have collaboration, which implies that using them together is optimal. To reduce the dropout and overfitting problems on a global scale, Bayesian optimization (BO) is a statistical optimization method.…”

Crop Recommendation and Monitoring using AI

“…The outcomes of the experiments show how well the suggested method works to correctly identify different plant diseases, which enhances agricultural management techniques. A cloud-enabled platform powered by machine learning algorithms for crop recommendation in precision farming is presented by Thilakarathne et al (2022) [2]. In an effort to maximize farming methods and raise output, the platform uses sensor data and cloud computing to give farmers customized crop advice.The authors create a machine learning-based method for crop recommendation based on sensor data, including soil composition, climate, and past yield information.…”

A Comprehensive Review on Deep Learning Assisted Computer Vision Techniques for Smart Greenhouse Agriculture

“…Similarly, using the PlantVillage dataset, Restrepo-Arias et al [104] introduced a novel diagnostic approach that highlights the impact of genotypic and phenotypic characteristics on how plants respond to pathogens. Their method, which emphasizes texture-based features and uses Bayesian Optimization to train artificial neural networks, achieved an impressive accuracy of up to 96.31% with MobileNet.…”

“…A major limitation identified is the shortage of large-scale standardized image datasets for the greenhouse domain (as noted in [65], [101], [104]). Most studies rely on small proprietary datasets collected by the researchers themselves, often just a few hundred images, which restricts generalization of techniques ( [92], [97], [106]).…”

“…2) Need for Specialized and Optimized Models While deep neural networks have driven progress in many visual recognition tasks, off-the-shelf models pre-trained on open-field images fail to account for the unique intricacies of greenhouse environments ( [95], [96], [104]). Networks designed for generic datasets do not transfer effectively to greenhouses, as the specialized appearance and growth patterns of crops in controlled settings are not represented ( [98], [107], [108]).…”