Intelligent Sine Cosine Optimization with Deep Transfer Learning Based Crops Type Classification Using Hyperspectral Images

Abstract

Hyperspectral Remote Sensing (HRS) is an emergent, multidisciplinary paradigm with several applications, which are developed on the basis of material spectroscopy, radiative transfer, and imaging spectroscopy. HRS plays a vital role in agriculture for crops type classification and soil prediction. The recently developed artificial intelligence techniques can be used for crops type classification using HRS. This study develops an Intelligent Sine Cosine Optimization with Deep Transfer Learning Based Crop Type Classification (ISCO-DTLCTC) model. The ISCO-DTLCTC technique comprises initial preprocessing step to extract the region of interest. The information gain-based feature reduction technique is employed to reduce the dimensionality of the original hyperspectral images. In addition, a fusion of 3 deep convolutional neural networks models namely, VGG16, SqueezeNet, and Dense-EfficientNet perform feature extraction process. Furthermore, sine cosine optimization (SCO) algorithm with Modified Elman Neural Network (MENN) model is applied for crops type classification. The design of SCO algorithm helps to proficiently select the parameters involved in the MENN model. The performance validation of the ISCO-DTLCTC model is carried out using benchmark datasets and the results are inspected under several measures. Extensive comparative results demonstrated the betterment of the ISCO-DTLCTC model over the state of art approaches with maximum accuracy of 99.99%. RÉSUMÉLa t el ed etection hyperspectrale (HRS) est une technologie emergente et multidisciplinaire ayant plusieurs applications d evelopp ees sur la base de la spectroscopie des mat eriaux, du transfert radiatif et de la spectroscopie des images. L'HRS joue un rôle essentiel en agriculture pour la classification des types de cultures et la pr evision des sols. Les techniques d'intelligence artificielle (IA) r ecemment d evelopp ees peuvent être utilis ees pour la classification des types de cultures a l'aide de HRS. Cette etude d eveloppe un mod ele intelligent d'optimisation du sinus-cosinus avec une classification des types de cultures bas ee sur l'apprentissage par transfert profond (ISCO-DTLCTC). La technique ISCO-DTLCTC comprend une etape initiale de pr etraitement pour extraire la r egion d'int erêt (RoI). La technique IGFR (Information Gain Based Feature Reduction) est utilis ee pour r eduire la dimensionnalit e des images hyperspectrales originales. Une fusion de trois mod eles DCNN (Deep Convolutional Neural Networks), a savoir VGG16, SqueezeNet et Dense-EfficientNet, effectue un processus d'extraction des principales caract eristiques. En outre, l'algorithme d'optimisation du sinuscosinus (SCO) avec le mod ele MENN (Modified Elman Neural Network) est appliqu e a la classification des types de cultures. La conception de l'algorithme SCO permet de s electionner efficacement les param etres impliqu es dans le mod ele MENN. La validation des performances du mod ele ISCO-DTLCTC est effectu ee a l'aide d'ensembles de donn ees de r ef erence et

A More Efficient Approach for Remote Sensing Image Classification

“…SSC via ML algorithms is an active research area in RSI data analysis. In fact, the critical step in almost all computer vision tasks is developing discriminative features to represent visual data [1][2][3][4][5][6]. In the past decades, the classical ML-SSC methods proposed in previous studies can be roughly divided into three categories according to the different feature extraction approaches, i.e., handcrafted features [7], unsupervised learning features [8], and DL features [9,10].…”