Joint Classification of Hyperspectral and LiDAR Data Based on Position-Channel Cooperative Attention Network

Round 1

Reviewer 1 Report

The authors proposed to use position-channel cooperative attention network to facilitate the joint classification of hyperspectral and LiDAR data. The method proposed was compared with other methods without the fused network, such as LiDAR, LiDAR-PC2A, HSI, HSI-PC2A, and was also compared with other advanced deep learning models such as Two-Branch CNN, EndNet, MDL-Middle, FusAtNet, IP-CNN, CRnN, S2ENet and HRWN. The results showed that the performance of PC2ANet is significantly better than methods without the fused network, and is slightly better than other advanced deep learning models. 

 

I would suggest the authors to elaborate on:

 

1. What is the training time cost for PC2ANet compared with the methods without the fused network and with other advanced deep learning models?

2. There are significantly more details in classified images using EndNet in Figure 9 and Figure 10. However, the OA, AA and Kappa parameters are lower than the other methods. Authors may want to explain why.  

3. Some minor issues:

   (i) Page 1: before HSI is used the first time, the full name should be given.  

   (ii) Page 2: 'foreign objects with the same spectrum' are repeated twice.

   (iii) line 210: 'output A, B, and C' are not indicated in Figure 3. 

   (iv) Line 222: superscript W misses 1.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Object classification is an important topic based on remote sensing. Due to the limitations  of feature representation, this paper  proposed  a PC2ANet to combine HSI and Lidar data, and the results proved the effectiveness of the Net. Overall, the article is well organized and its presentation is good. However, some issues still need to be improved. (1) the abstract  and key words should be more clear to focus  on the innovations. (2) the introduction and related work is not well organized. (3) the method should be clear. (4) the references need to be simplified. (5) the manuscript need to be carefully editing.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

In this interesting manuscript the authors  propose HSI and LiDAR data classification using position-channel cooperative attention network. Multiscale networks and self-position enhancement attention networks are used to extract deep HSI features. Then, a forward-inverted CNN structure is designed to extract rich spatial features from LiDAR images and fused with extracted HSI features to obtain spatial-spectral information.

The article overall seems well written and worthy of publication. However, I suggest a re-reading to eliminate the very few remaining imperfections, as:

lines 7 and 11: Please, avoid acronyms in the abstract unless the acronym is used multiple times in the abstract. If an acronym is used in the abstract, it must be defined in the abstract, and then spelled out again the first time it is used in the body of the paper.

line 27: Please, add the explanation of the acronym the first time it is mentioned in the introduction.

line 205: Figures should be referred into the text before being placed.

line 335: Respecting correct formatting, the caption of Figure 8 should be well separated by this line.

To complete the literature references, assess whether doi: 10.1109/TGRS.2013.2292544 can be added to the references, given the similarity of topics.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf