Multi-temporal grass hyperspectral classification via full pixel decomposition spectral manifold projection and boosting active learning model

ABSTRACT

Grassland detection is essential for sustainable development, particularly in ecological and animal husbandry. Hyperspectral can retrieve the functional characteristics of vegetation species, nevertheless, the efficiency of grassland communities with rich species at different periods needs to be improved. In this study, the multi-temporal grass hyperspectral classification via full pixel decomposition spectral manifold projection and boosting active learning (FPD-SMP-BAL) model is proposed to relieve the above issue. Considering the plentiful spectral information of hyperspectral, we proposed a novel preprocessing method of spectral manifold projection based on full pixel decomposition (FPD-SMP), which reveals the internal nonlinear structure of global hyperspectral and alleviate the data redundancy. Furthermore, we innovatively designed the boosting active learning (BAL) classification model with an effective optimization strategy, which can decrease the cost of sample labeling by the minimal loss value. An on-site hyperspectral acquisition platform was assembled, and a dataset of 4,200 samples was established from the perspective of vegetation species composition and multi-stages to verify the validity of the model. Experimental results show that the proposed FPD-SMP-BAL framework reaches an overall accuracy of 94.94%, which is superior to some state-of-the-art methods in the classification of multi-temporal grass hyperspectral images, and also provides a reference for related research of grassland dynamic monitoring.

KEYWORDS:

• Hyperspectral classification
• Multi-temporal
• grass
• Manifold structure
• Active learning

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

The hyperspectral data that support the findings of this study are available from the corresponding author, Pan X., upon reasonable request.

Additional information

Funding

This work was supported by National Natural Science Foundation of China (Grant No. 61962048, Grant No. 61562067, Grant No. 52079063), Technological Achievements of Inner Mongolia Autonomous Region of China (Grant No. 2020CG0054), Natural Science Foundation of Inner Mongolia Autonomous Region of China (Grant No. 2019JQ06), Program for Innovative Research Team in Universities of Inner Mongolia Autonomous Region (Grant No. NMGIRT2313), Central Public-interest Scientific Institution Basal Research Found (Grant No. 1610332020020), and the Center of Information and Network Technology of Inner Mongolia Agricultural University.