Homomorphic Integral and Dual Rectified Convolutional Grading for Salinity Prediction During Seedling in Rice

Abstract

Rice is considered as one of the most sought out food crops globally. Salinity resilience plays an essential part in rice cultivation. But, uncontrolled saline level specifically at the seedling stage may unfavorable influence productivity of rice extensively. If not properly monitored, salinity stress brings about acute impairment during the seedling stage of crop growth bringing about an overall loss of 50%. Hence, it is mandatory to design an image classification and grading learning method for salinity stress analysis principally at the seedling stage and then benchmark its performance to circumvent depletion in rice yield. However, the classification using traditional method is both said to be laborious in terms of time and most of the time results in error owing to incorrect classification. To identify and classify salinity stress in rice seedlings using field images, the study reports the need of deep learning built model over traditional method of assessing rice crop's susceptibility to salt stress during the seedling stage. In this work, a method called Homomorphic Fourier Integral-based image classification and Dual Rectified Linear Convolutional (HFI-DRLC) grading learning for salinity stress level analysis in rice crops at seedling stage is proposed. Initially, an image enhancement framework that balances both field image features like as contrast, illumination and key properties that are important for identification of salinity stress level in rice crops at seedling stage. This is performed using Homomorphic Fourier Integral Filter-based Preprocessing model. Next, Dual Rectified Linear Convolutional Neural prediction-based salinity tolerance in rice at seedling stage is designed to ensure both accuracy and precision. To investigate the effects of different improvement methods on the accuracy and precision of salinity tolerance in rice at seedling stage, comparison experiments between HFI-DRLC and traditional methods and comparison experiments with rice seedling dataset and rice seedling samplings from ICAR (Central Coastal Agricultural Research Institute, Old Goa, India) are executed. The method proposed in this study was shown to be more effective than traditional methods in terms of precision, recall, accuracy and error rate, providing a better method for salinity tolerance in rice at seedling stage.

Plant phenotyping relevance

イネ幼苗の塩ストレス状態を圃場画像から分類・等級化する画像処理および深層学習手法を開発し、従来法やデータセットで性能比較しており、表現型取得・推定が研究の中心である。

Code and data availability

The paper evaluates its HFI-DRLC salinity-grading method on a public rice seedling image dataset hosted on figshare, explicitly cited with URL, plus a non-public real-time ICAR sample set. The figshare dataset is a public plant-image asset directly used for the paper's salinity phenotyping analysis. No author analysis,