Sea Cucumbers

The animal material for genome sequencing and assembly was from a male A. japonicus captured off the coast of Laoshan, Qingdao, China. The sea cucumber was acclimated in sea water at 15 ± 1°C before experiments. Muscle, gonad, and respiratory tree tissues were collected and immediately frozen in liquid nitrogen and stored at −80°C. Genomic DNA was extracted using a TIANamp Marine Animal DNA Kit (TIANGEN, Beijing, China) according to the manufacturer’s instructions.

To identify neuropeptide precursors in O. victoriae, A. filiformis and O. aranea, sequences of neuropeptide precursors identified previously in other echinoderms (including the starfish, A. rubens, the sea urchin S. purpuratus and the sea cucumber, A. japonicus) were used as queries for tBLASTn analysis of a transcriptome database, using an e-value of 1000. Sequences identified as potential neuropeptide precursors by BLAST were translated using the ExPASy Translate tool (http://web.expasy.org/translate/) and then analysed for features of neuropeptide precursors. Specifically, sequences were evaluated based on (i) the presence of an N-terminal signal peptide (using Signal P v. 4.1 with the sensitive cut-off of 0.34) and (ii) the presence of monobasic or dibasic cleavage sites flanking the putative bioactive peptide(s).
To identify novel neuropeptide precursors or highly divergent precursors with low sequence similarity to known precursors, we used two additional approaches. In the first approach, we used NpSearch [8 (link)], software that identifies putative neuropeptide precursors based on various characteristics (presence of signal peptide and dibasic cleavage sites among others). In the second approach, NpHMMer (http://nphmmer.sbcs.qmul.ac.uk/), an HMM-based software, was used to identify neuropeptides not found using the above approaches.
Neuropeptide precursors identified in O. victoriae (which represented a more comprehensive neuropeptide precursor repertoire compared to A. filiformis and O. aranea) were then submitted as queries for BLAST analysis of sequence data from 52 Ophiuroidea species, using an E-value of 1 × 10⁻⁶. BLAST hits were then further analysed using an automated ruby script (available at https://github.com/IsmailM/ophiuroid_neuropeptidome). Each BLAST hit was translated using BioRuby, and the open reading frame (ORF) containing the BLAST high-scoring segment pair (HSP) was extracted. These ORFs were then examined for the presence of a signal peptide using Signal P 4.1 using a sensitive cut-off of 0.34. All sequences were then aligned using MAFFT, with the number of maximum iterations set to 1000 to ensure an optimal alignment. These alignments were then further optimized by manually adjusting the location of the bioactive peptide and cleavage sites. Finally, the alignments were annotated using different colours for the signal peptide (blue), the bioactive peptide(s) (red) and cleavage sites (green).

Common metrics of object detection are adopted to evaluate the performance of sea cucumber detection. The individual detection results are evaluated and compared by the precision–recall analysis and average precision (AP) [36 (link),37 (link)]. The precision–recall analysis is conducted by calculating true positive (TP), false positive (FP) and false negative (FN). True positives are the detected results which match the ground truth. False positives are the results reported by the detection algorithm but are actually incorrect. In other words, TP contains sea cucumber individuals, whereas FP has no sea cucumbers. Usually, background objects are confused with the detection targets due to appearance similarity or inaccurate detectors. In object detection, the target objects which cannot be identified by the detection algorithm are counted as false negatives. The precision measures the proportion of correct results from the total detection results. High precision indicates the detection results containing a high percentage of reliable results and a low percentage of false alarms. Precision is calculated from TP and FP (Equation (1)). Recall represents the detection accuracy of sea cucumbers and refers to the percentage of correctly detected individuals from the total number of sea cucumbers (Equation (2)). To evaluate the overall performance of object detection, F measure is calculated by considering both precision and recall (Equation (3)). A high F measure score indicates that the detection results are accurate and reliable.
$$\mathrm{Precision}=\frac{\mathrm{True} \mathrm{Positive}}{\mathrm{True} \mathrm{Positive}+\mathrm{False} \mathrm{Positive}}$$
$$\mathrm{Recall}=\frac{\mathrm{True} \mathrm{Positive}}{\mathrm{True} \mathrm{Positive}+\mathrm{False} \mathrm{Negative}}$$
$$\mathrm{F}=\frac{2\cdot \mathrm{Precision}\cdot \mathrm{Recall}}{\mathrm{Precision}+\mathrm{Recall}}$$
Average precision (AP) is a widely applied metric for evaluating object recognition/detection [37 (link)]. Average precision calculates the shape of the precision/recall curve and replaces the area-under-curve (AUC) of ROC curve to improve the sensitivity of the metric. Average precision is defined as the mean precision at 11 equally divided levels in recall [0, 0.1, …, 1]. The calculation of AP is given in Equation (4): $$\mathrm{AP}=\frac{1}{11}{\displaystyle \sum _{r\in \left\{0,0.1,\dots ,1\right\}}{P}_{interp}(r)}$$
where P_interp represents the interpolated precision at a certain recall level r. Another term for evaluating detection accuracy is IOU. This measures the area of overlap a_o between the detected bounding box B_p and ground truth bounding box B_gt. Intersection over union evaluates the accuracy of predicted bounding box: $$a_o=\frac{area\left(B_p\cap B_{gt}\right)}{area\left(B_p\cup B_{gt}\right)}$$
where B_p ∩ B_gt is the intersection of the two bounding boxes and B_p ∪ B_gt is the union of them. Usually, a threshold of 50% of IOU is required to examine the detection results.

In general, image enhancement is utilized to emphasize the global or local features of an image, such as improving the color representation, brightness and contrast of an object. Image enhancement is widely applied to improve the clarity of images, emphasizing certain features of interest, enlarging the differences between objects and backgrounds and suppressing uninteresting features. Therefore, image enhancement has been commonly used in underwater image processing and target detection. In this work, five well-known image enhancement methods are chosen to evaluate the efficiency of image enhancement in underwater sea cucumber detection. These image enhancements are contrast limited adaptive histogram equalization (CLAHE), dark-channel prior (DCP), non-local image dehazing (NLD), Retinex, and underwater generative adversarial network (UGAN), which covers histogram-based method, dehazing method, physical color model, and deep learning-based method.
Contrast limited adaptive histogram equalization is a variant of adaptive histogram equalization (AHE). CLAHE can reduce the noise problem of AHE by limiting contrast enhancement [32 ]. It calculates multiple histograms and each of them corresponds to a different part of the image. The brightness of the image is redistributed according to these histograms. CLAHE limits the amplification by clipping the histogram at a user-defined value called clip limit. The clipping level determines how much noise in the histogram should be smoothed and hence how much the contrast should be enhanced. Thus, CLAHE is suitable for enhancing the local image contrast and emphasizing edge features in each part of the image.
Dark-channel prior is a statistical rule for haze-free images. He et al. found that there are always pixels with at least one intensity value that is close to zero within an image patch [33 (link)]. In the process of dark channel extraction, the image is decomposed in RGB space, and the minimum value operation is taken in the local block to obtain the minimum component in the three channels (R, G, B). A Marcel Van Herk’s algorithm is used to implement the local region minimum filtering on the minimum component value, i.e., the gray level corrosion operation. The effectiveness of DCP in dehazing is proved by its applications in solving haze removal issues.
Non-local image dehazing assumes that colors of a haze-free image are well approximated by a few hundred distinct colors that form tight clusters in RGB space and pixels in a cluster are often non-local [34 ]. The term haze-line is proposed to estimate the transmission factors. In this method, clustering is used to group the pixels so that each cluster becomes a haze-line. Then, the maximum radius of each cluster is calculated and used to estimate the transmission. A final regulation step is performed to enforce the smoothness of the transmission map. The NLD could improve the visibility and enhance the detailed image features.
Retinex is a composite of retina and cortex and is referred to as the retinal cortex theory. The basic idea of Retinex theory is that the illumination intensity determines all pixels in the original image, and the inherent property of the original image is determined by the reflection coefficient of the object itself. That is, the reflection image and the illumination image are assumed to be the original image. Therefore, Retinex is to remove the influence of illumination and retain the inherent property of the object [35 ].
Recently, the generative adversarial network (GAN) presented outstanding performance in image synthesis and style transferring. The underwater GAN (UGAN) uses an adversarial approach towards generating realistic underwater images. UGAN structures the problem of estimating the real appearance of underwater imagery as a paired image-to-image translation problem [26 ]. In the training process, UGAN learns the restoration model from the image pairs taken in two independent domains (e.g., underwater and ground).
Examples of enhanced images of our datasets are illustrated in Figure 7. The selected image enhancement methods present various characters of images in different scenes. The enhanced image sets are used to train the object detection models and to evaluate the efficiency of each enhancement method for detecting sea cucumbers.