Attention Mechanisms in Convolutional Neural Networks for Nitrogen Treatment Detection in Tomato Leaves Using Hyperspectral Images

Abstract

Nitrogen is an essential macronutrient for the growth and development of tomatoes. However, excess nitrogen fertilization can affect the quality of tomato fruit, making it unattractive to consumers. Consequently, the aim of this study is to develop a method for the early detection of excessive nitrogen fertilizer use in Royal tomato by visible and near-infrared spectroscopy. Spectral reflectance values of tomato leaves were captured at wavelengths between 400 and 1100 nm, collected from several treatments after application of normal nitrogen and on the first, second, and third days after application of excess nitrogen. A new method based on convolutional neural networks (CNN) with an attention mechanism was proposed to perform the estimation of nitrogen overdose in tomato leaves. To verify the effectiveness of this method, the proposed attention mechanism-based CNN classifier was compared with an alternative CNN having the same architecture without integrating the attention mechanism, and with other CNN models, AlexNet and VGGNet. Experimental results showed that the CNN with an attention mechanism outperformed the alternative CNN, achieving a correct classification rate (CCR) of 97.33% for the treatment, compared with a CCR of 94.94% for the CNN alone. These findings will help in the development of a new tool for rapid and accurate detection of nitrogen fertilizer overuse in large areas.

Hyperspectral Image Shadow Enhancement Using Three-Dimensional Dynamic Stochastic Resonance and Classification Based on ResNet

“…Attention modules have been shown to enhance the feature extraction and generalization performance of neural networks [39,40]. CBAM can be directly embedded into current mainstream CNN network structures, enhancing the feature extraction capabilities without significantly increasing computational and parameter requirements.…”

Discriminating Spectral–Spatial Feature Extraction for Hyperspectral Image Classification: A Review

“…SK modules and FCA modules enhanced the effectiveness of spectral attention with different scales of features [156][157][158] and frequency features [166][167][168], respectively. By introducing adaptive average pooling [160] and global max-pooling [161][162][163][164], PA modules and spe-CBAMs perceived different scales of contextual information and global salient responses, separately. These attention modules have become powerful means to capture salient bands for discriminating spectral features.…”

“…GE modules [169] utilize depth-wise convolution to gather and assess the correlations between spectral features in small regions and resize aggregated weights for adjustment [170][171][172]. To consider more useful information of input, spa-CBAMs [147] introduce global average pooling and max-pooling layers before convolution, which improve spatial attention without increasing the number of parameters [161][162][163][164][173][174][175]. Different from GE modules and CBAMs, 1 × 1 × 1 convolutional layers were exploited in BAMs [176] to compress and transform the information in spectral and channel dimensions [177][178][179], which enhanced the adaptation of spatial attention.…”

“…The resultant spectral attention depends on global interactions between all bands, but local interactions may be neglected. To solve this problem, many extensions of the SE module, including efficient channel attention (ECA) modules [152][153][154], selective kernel (SK) modules [155][156][157][158], pyramid attention (PA) modules [159,160], spectral part of CBAMs (spe-CBAMs) [147,[161][162][163][164], and frequency channel attention (FCA) modules [165][166][167][168], were designed to strengthen the relationships between various types of crucial information and spectral attention. This mode of global information transformation has been proven to be effective in locating the distinctive bands.…”