Logged Sites Research Articles

Evaluation of Ammonia Co-fuelling in Modern Four Stroke Engines

Ammonia is emerging as a promising alternative fuel for longer range decarbonised heavy transport, particularly in the marine sector due to highly favourable characteristics as an effective hydrogen carrier. This is despite generally unfavourable combustion and toxicity attributes, restricting end use to applications where robust health and safety protocols can be upheld. In the currently reported work, a spark ignited thermodynamic single cylinder research engine equipped with gasoline direct injection was upgraded to include gaseous ammonia port injection fuelling, with the aim of understanding maximum viable ammonia substitution ratios across the speed load operating map. The work was conducted under overall stoichiometric conditions with the spark timing re-optimised for maximum brake torque at all stable logged sites. The experiments included industry standard measurements of combustion, performance and engine-out emissions (including ammonia ‘slip’). With a geometric compression ratio of 12.4:1, it was possible to run the engine on pure ammonia at low engine speeds (1000–1800 rpm) at low to moderate engine loads in a fully warmed up state. When progressively dropping down below a threshold load limit, an increasing amount of gasoline co-firing was required to avoid engine misfire. Due to the favourable antiknock characteristics, pure ammonia operation was up to 5% more efficient than pure gasoline operation under stable operating regions. A maximum net indicated thermal efficiency (ITE) of 40% was achieved, with efficiency tending to increase with speed and load. For the co-fuelling of gasoline and ammonia in a pure ammonia attainable operating region, it was found that addition of gasoline improved the combustion, but these improvements were not sufficient to translate into improved thermal efficiency. Emissions of ammonia slip reduced with increased gasoline co-fuelling, albeit with increased NOx. However, the reduction in ammonia slip was nearly ten times the increase in NOx emissions. Comparing pure ammonia and pure gasoline operation, NOx reduced by ~60% when switching from pure gasoline to pure ammonia (as the latter is associated with longer and cooler combustion). Results were finally compared to those obtained a modern multicylinder Volvo ‘D8’ turbo-diesel engine modified for dual-fuel operation with ammonia port fuel injection (PFI), with the focus of the comparison being ammonia slip and NOx emissions.

Public Catering at Logging Sites in the Middle Volga Region during the Pre-War Five-Year Plans

The article examines one of the most important aspects of the industrial and domestic components of the logging process in the 1930s. The work of canteens and food stations for lumberjacks is considered using the example of the Middle Volga region, whose logging was typical for the European part of the USSR. At the same time, the article compares regional problems of catering at logging sites with all-Union trends. The source base for the article was the funds of regional archives of the Penza, Samara and Ulyanovsk regions and materials from periodicals of the 1930s. In general, the catering system at the logging sites of the largest logger in the Middle Volga region, the Sredles trust, was organized already at the end of the first five-year plan. The greatest number of problems in the public catering system in the third five-year plan were recorded in the Kuibyshevles sections. Typical methods for supplying the canteens of timber industry enterprises with products in the 1930s. began to organize their own subsidiary farms and purchase food. There are numerous complaints about the quality of food prepared in the forestry industry canteens. In some cases, the shortcomings of catering at logging sites were associated not with the general food shortage in the USSR, but with the low professionalism of canteen workers.

A conditioned Latin hypercube sampling design methodology for ground-truthing transient EM resistivity models

Translating 3D resistivity models of the subsurface to relevant hydrogeological parameters such as sediment texture requires applying a rock physics transform that is ideally calibrated using co-located 1D sediment texture logs. It is difficult to select optimal logging sites in the model that span the range of observed resistivity values using expert opinion alone. Instead, we demonstrate a methodology for selecting logging sites using conditioned Latin hypercube sampling (cLHS), a maximally stratified sampling design methodology. cLHS employs simulated annealing to select sampling sites that minimize an objective function by reproducing the statistics of the resistivity model. We implement numerous preprocessing steps to isolate desirable regions within the model for logging, and we use principal component analysis to reduce the dimensionality of the model to aid the convergence of the simulated annealing procedure while preserving the majority of the model variance. Additionally, we impose an additional objective function condition to penalize the selection of sites in the resistivity model that exhibit a high degree of lateral variance. This condition is used to increase the likelihood that the sediment texture logs acquired at the selected sites correlate well with the resistivity model. We tested our approach on a resistivity model obtained using a towed transient EM system in an almond orchard in California’s Central Valley, comparing our selected sets of logging sites against the five cone penetrometer testing logs originally acquired at the site, as well as randomly selected logging sites. The sample sets selected with cLHS were consistently able to reproduce the statistics of the resistivity model and did so more effectively than the true sampling locations or the random sampling locations.

Tree Species–Soil Nutrients Association in Unlogged and Logged-over Tropical Forests, Sabah, Borneo

Natural forest cover in Sabah has been decreasing across all forest types for more than two decades since the 1990s, being replaced mostly by man-made forests. This contributes to the overall degradation of the forest ecosystem, which changes soil nutrient content and subsequently modifies forest species composition, as well as vegetation structure. This study investigated the association between soil nutrients and tree species in unlogged and logged tropical lowland forests in Sabah, Borneo, Malaysia. A total of eight and 27 sampling plots were established in randomly placed locations at the Danum Valley Conservation Area representing the unlogged site, as well as the Sapat Kalisun area of the Ulu Segama Forest Complex representing the logged site. Soil samples at two depths (0–5 cm; 5–10 cm) were collected from each plot and analyzed for soil nutrients: total N, available P, inorganic N (NO3 – and NH4 +), exchangeable cations (K+, Ca2+, Mg2+, and Na+), exchangeable Al3+, and exchangeable acidity. Results showed that certain exchangeable cations and soil nutrient concentrations were greater in the unlogged forest compared to the logged forest except for K+, Na+, and inorganic N. Greater inorganic N concentrations were observed in the logged forest, hence suggesting a dominant shift toward inorganic N in the soil. PCA results indicated that exchangeable acidity and exchangeable Al3+ were correlated to logged forests, whereas Ca2+ and Mg2+ were associated with unlogged forests. CCA ordination showed the existence of association between specific tree species with K+, Ca2+, Mg2+, and Na+ in the unlogged forest only. Logging, to some extent, influenced the status of certain soil nutrients but their association with tree species was complex. This could be attributed to the heterogeneous habitat resulting from past logging disturbances.

Logging Machinery Traffic Has Greatest Influence on Soil Chemical Properties in the Amazonian Rainy Season

Abstract In the forests of Amazonia each year, previously unentered stands are logged, which usually results in some degree of soil compaction. Consequently, the soil chemical properties in compacted areas are altered. The aim of this study was to determine how these changes may help or hinder site recovery in the context of seasonal variation and increased levels of compaction. To investigate these changes, an experiment was established in the Central Amazon. This consisted of tractor trails compacted at three incrementally increasing traffic intensities of one, three, and twelve machine cycles in the wet and dry seasons. Results revealed that elevated moisture in the wet season combined with heavy compaction from twelve machine cycles had the greatest impact on soil chemical properties. This was indicated by diminished total nitrogen, organic carbon, available phosphorus, cation exchange capacity, and elevated ammonium and Fe2+. Nevertheless, heavy compaction in the wet season led to lower Al3+ and higher base saturation, which could be beneficial for future site recovery. Study Implications: Annually, vast swathes of old-growth forest are logged for the first time in Amazonia. These areas require an extensive skid trail network to facilitate movement of logging machinery and subsequent log skidding. Skid trails always entail some level of soil compaction, especially in the heavily used primary skid trails. This compaction influences the soil chemical properties. Some impacts are only apparent in the wet season and may not be present in the dry season. Therefore, to reduce overall impacts to soil chemical properties, skid trail coverage should be planned to reduce the coverage of the logging site.

A novel gradient descent optimizer based on fractional order scheduler and its application in deep neural networks

To improve convergence speed of deep neural network trained by gradient descent methods (GDMs), this paper proposed a novel fractional order gradient descent optimizer (FOAdam) based on Caputo operator and Adam algorithm, and analyzed the impact of fractional order on convergence performance using optimization theory and simulation. Furthermore, to address the trade-off between convergence speed and precision under fixed-order conditions, a fractional order scheduler (FOS) was designed based on the Connections Cloud Model (CCM). Numerical experiments conducted on MNIST and CIFAR-10 datasets showed that FOS-FOAdam outperformed other mainstream GDMs in terms of convergence speed and precision. Finally, FOS-FOAdam was applied in recognizing and classifying image datasets of cuttings and cores at the logging site, demonstrating good prospects for practical engineering applications.

Investigation of the Process of Obtaining Flour from Wood Greens of Coniferous Species

An analysis of the waste generation process from wood during logging operations is presented in this paper. The authors presented the concept of a device that can produce flour from green coniferous wood. An evaluation was also conducted regarding the quantity of waste derived from logging residues obtained directly from cutting sites, as well as the potential areas for utilization. The use of coniferous wood waste as a potential raw material for the production of various types of products is also noted. The concept of a mobile knife installation has been proposed for the generating flour from the greenery of coniferous trees, directly at the logging site. A multifactorial experiment was planned and implemented in order to determine the optimal design and technological parameters for the proposed installation and design of an industrial device. The investigation was conducted utilizing both analytical and experimental techniques. The use of a mobile installation for the production of flour from green coniferous wood, in conditions of logging operations, will expand the raw material base for the manufacture of products using pine needles, as well as reduce the amount of waste generated during the wood harvesting process. It will have a positive impact not only on the ecological situation of the environment, but also on the likelihood of a forest fire occurring and spreading.

Evaluation of ammonia-gasoline co-combustion in a modern spark ignition research engine

Ammonia (NH3) is emerging as a potential favoured fuel for longer range decarbonised heavy transport, particularly in the marine sector, predominantly due to highly favourable characteristics as an effective hydrogen carrier. This is despite generally unfavourable combustion and toxicity attributes, restricting end use to applications where robust health and safety protocols can always be upheld. In the currently reported work, a spark ignited thermodynamic single cylinder research engine equipped with gasoline direct injection was upgraded to include gaseous ammonia port injection fuelling, with the aim of understanding maximum viable ammonia substitution ratios across the speed-load operating map. The work was conducted at varied effective compression ratios under overall stoichiometric conditions, with the spark timing re-optimised for maximum brake torque at all stable logged sites. The experiments included industry standard measurements of combustion, performance, and engine-out emissions (including NH3 “slip”). With a geometric compression ratio of 11.2:1, it was found possible to run the engine on pure ammonia at low engine speeds (1000-1800 rpm) and loads of 12 bar net IMEP. When progressively dropping down below this load limit an increasing amount of gasoline co-firing was required to avoid engine misfire. When operating at 1800 rpm and 12 bar net IMEP, all emissions of carbon (CO2, CO, unburned hydrocarbons) and NOx decreased considerably when switching to higher NH3 substitution ratios, with NOx reduced by ~ 45% at 1800 rpm/12 bar when switching from pure gasoline to pure NH3 (associated with longer and cooler combustion). By further increasing the geometric compression ratio to 12.4 and reducing the intake camshaft duration for maximum effective compression ratio, it was possible to operate the engine on pure ammonia at much lower loads in a fully warmed up state (e.g., linear low load limit line from 1000 rpm/6 bar net IMEP to 1800 rpm/9 bar net IMEP). Under all conditions, the indicated thermal efficiency of the engine was either equivalent to or slightly higher than that obtained using gasoline-only due to the favourable anti-knock rating of NH3. Ongoing work is concerned with detailed breakdown of individual NOx species together with measuring the impact of hydrogen enrichment across the operating map.

Linkages of soil CO2 emission with plant functional traits in young subtropical plantations.

Soil respiration is a key process in forest biogeochemical cycling. Exploring the relationship between plant functional traits and soil respiration can help understand the effects of tree species conversion on soil carbon cycling. In this study, we selected 15 common subtropical tree species planted in the logging site of second-generation Chinese fir forest to measure soil CO2 emission fluxes, soil physicochemical properties, leaf and root functional traits of each species, and explored the effects of plant functional traits on soil respiration. The results showed that the annual flux of soil CO2 emissions varied from 7.93 to 22.52 Mg CO2·hm-2, with the highest value under Castanopsis carlesii (22.52 Mg CO2·hm-2) and the lowest value under Taxus wallichiana (7.93 Mg CO2·hm-2). Results of stepwise regression analysis showed that the annual flux of soil CO2 emission decreased with the increases of leaf nitrogen content and fine root diameter, and increased with increasing leaf non-structural carbohydrate. In the structural equation model, leaf non-structural carbohydrate had a direct and significant positive effect on soil CO2 emission fluxes, while leaf nitrogen content and fine root diameter had a direct negative effect by decreasing soil pH and soluble organic nitrogen content. Plantations of different tree species would affect soil CO2 emission directly by changing functional traits related to water and nutrient acquisition or indirectly through soil properties. When creating plantations, we should select tree species based on the relationship between plant functional traits and ecosystem functions, with a view to improving forest productivity and soil carbon sequestration potential.

Data-driven wireline sticking risk assessment and control factor analysis

The risk of wireline sticking cannot be accurately predicted at wireline logging site. Once the wireline is sticked on the wellbore, it seriously affects the logging process. A data-driven risk assessment method has been proposed to quantitatively predict and qualitatively evaluate the probability of wireline sticking accidents. Firstly, based on the data collection capability and expert experience, the risk characterization parameters of wireline sticking were determined. Secondly, according to the operating conditions of wireline logging, the variable length window data acquisition method and risk calibration were established, and the data feature extraction and feature selection methods were established. Thirdly, 149 logging operation records were collected to establish training data, and synthetic minority over-sampling technique (SMOTE) algorithm were used to expand the number of training data. Finally, a probability prediction model based on co-training-style random forest (Co-Forest) was established to quantitatively predict the probability of cable jamming. And the critical risk probability value is observed by fuzzy clustering to define the risk level. The performance of the proposed method was verified by logging operation records from an oilfield from 2019 to 2022, and the prediction accuracy was more than 90%. The results show that the model has good guiding significance and practicability for realizing intelligent wireline logging operation safety warning. The proposed method can be used to determine the depth of wireline logging operations and avoid wireline sticking.

Towards the use of satellite-based tropical forest disturbance alerts to assess selective logging intensities

Illegal logging is an important driver of tropical forest loss. A wide range of organizations and interested parties wish to track selective logging activities and verify logging intensities as reported by timber companies. Recently, free availability of 10 m scale optical and radar Sentinel data has resulted in several satellite-based alert systems that can detect increasingly small-scale forest disturbances in near-real time. This paper provides insight in the usability of satellite-based forest disturbance alerts to track selective logging in tropical forests. We derive the area of tree cover loss from expert interpretations of monthly PlanetScope mosaics and assess the relationship with the RAdar for Detecting Deforestation (RADD) alerts across 50 logging sites in the Congo Basin. We do this separately for various aggregation levels, and for tree cover loss from felling and skidding, and logging roads. A strong linear relationship between the alerts and visually identified tree cover loss indicates that with dense time series satellite data at 10 m scale, the area of tree cover loss in logging concessions can be accurately estimated. We demonstrate how the observed relationship can be used to improve near-real time tree cover loss estimates based on the RADD alerts. However, users should be aware that the reliability of estimations is relatively low in areas with few disturbances. In addition, a trade-off between aggregation level and accuracy requires careful consideration. An important challenge regarding remote verification of logging activities remains: as opposed to tree cover loss area, logging volumes cannot yet be directly observed by satellites. We discuss ways forward towards satellite-based assessment of logging volumes at high spatial and temporal detail, which would allow for better remote sensing based verification of reported logging intensities and tracking of illegal activities.

Stem segmentation for sustainable forest management task

Harvesting of forest raw materials, namely felling and bucking of log on forest areas, is the first and main stage in the logging chain. One of the problems in this industry is the shortage of highly qualified specialists-operators of forest machines, including feller-delimbing-bucking machines (harvesters). Operators who have just come to the industry or have insufficient experience (have worked for less than a year) cannot correctly configure harvesters, as a result of which the processes in the logging chain are disrupted. Thus, it becomes necessary to apply additional models of working with operators and forest machines to reduce the resulting costs for the company. Understanding how much wood raw material will be obtained from the logging site allows predicting not only the amount of equipment required, but also planning actions for the next stages of logging.

Average Skidding Distance of Timberin Non-Rectangular Cutting Areas

The results obtained in this article relate to the scientific and technological development of the timber industry. The main idea of the article is based on the possibility of concentrating several forest plots designated for felling with different percentages of cut components within a single cutting area. On the logging site created in this way, single main skidding and skidding trails were laid, allowing the entire range of logging operations to be carried out. The shape of such combined cutting areas can be a complex geometric figure. The purpose of this article is to improve existing methods and mathematical dependencies to justify the average skidding distance of timber when combining adjacent forest areas into a single cutting area. The article uses the methods of mathematical analysis, in particular, the methods of differential and integral calculus. The developed algorithm made it possible to create a unified methodological approach to find the average skidding distance, regardless of the shape of the cutting area. An important distinguishing characteristic of the created mathematical dependencies is that they can be used in cutting areas consisting of several forest plots characterized by different wood stocks. The forest areas included in the cutting area may differ in the presence of a branched structure of the main skidding and skidding trails. The results obtained can be used effectively by logging enterprises in substantiating the production rates and planning the work for future periods. As a result of using the obtained mathematical dependencies, it becomes possible to effectively place loading points and skidding trails on the territory of the cutting area.

Размерно-качественные характеристики круглых сортиментов, заготавливаемых с применением многооперационных лесных машин

It is necessary to carry out a statistical assessment of the harvested wood parameters in order to increase the yield of round timber and the overall productivity of logging operations while operating multi-operational forest machines. Modern forestry harvesters are equipped with a variety of CAN control sensors to monitor operation and have onboard software that can collect and store input data on various file types to ensure continuous correct operation. One of the main file types is STM (trunk files). These files are presented in two forms: some contain information for each individual tree trunk in different files, and others store all the data for all trunks in one. The second type of STM files is the most relevant when analyzing the operations carried out at the logging site. However, it is inconvenient to use STM files every time to get information, so there is a need to convert the initial data into a table for easy transfer and decision-making at risk and uncertainty. The article proposes a method for transferring such data to a CSV table using Python pandas, numpy, seaborn, matplotlib programming language libraries, which help to process large data arrays quickly and efficiently and display them graphically. The data obtained by two operators using a medium-class Ponsse Ergo 8W machine in the typical natural and production conditions of the Mondi Syktyvkar JSC (Komi Republic, middle taiga zone) were used for the transfer. The efficiency of the operators’ work was assessed. We obtained functions for determining the volume of a tree trunk on the basis of the reported data from the forest machines. The analysis of structured data on the operation of multi-operational forest machines helps to improve decision-making during subsequent felling of trees with the selection of species, which provides the largest volume of round timber output. Moreover, it is possible to adjust assortment tables (APT-matrices) for shortand medium-term planning of assortment harvesting volumes.

Use of Hydrological Models to Predict Risk for Rutting in Logging Operations

Using hydrological models with a high temporal resolution to predict risk for rutting may be a possible method to improve planning of forwarder trails or to schedule logging operations in sites with low bearing capacity to periods when soil moisture content is at a minimum. We have studied whether descriptions of rut variations, collected in 27 logging sites, can be improved by using hydrological data, modeled by Swedish HYdrological Prediction for Environment (S-HYPE). Other explanatory variables, such as field-surveyed data and spatial data, were also used to describe rut variations within and across logging sites. The results indicated that inclusion of S-HYPE data led to only marginal improvement in explaining the observed variations of the ruts in terms of both “rut depths” within the logging sites and “proportion of forwarder trails with ruts” across the logging sites. However, application of S-HYPE data for adapting depth-to-water (DTW) maps to temporal changes of soil moisture content may be a way to develop more dynamic soil moisture maps for forestry applications.

Модернизация валочно-сучкорезно-раскряжевочного механизма лесного харвестера

There is an annual increase in the number of modern forestry machines at logging enterprises both in Russia and in foreign countries. At the same time, the fleet of forestry machines is becoming more diverse and multifunctional, and their manufacturers, competing for market outlets, find themselves interested in constant upgrading of existing and development of new design and technological solutions. The research is aimed at finding a new engineering solution that will allow along with harvesting high-quality logs to perform their partial debarking without a significant reduction in the processing speed. The effective result is achieved by the fact that the proposed new method of processing tree trunks provides the possibility of timber barking in strips, i.e. removing the bark from a fallen tree along its entire length, as well as bast removal (partially or completely). The method involves, as in the traditional model of harvesting (processing) head functioning, crosscutting of the trunk by the sawing mechanism and delimbing by delimbing knives. This leads to accelerated natural drying of wood during its storage in the warehouses and significantly increases the floatability of thin coniferous logs during their further timber floating. The implementation of the proposed option of harvesting debarked timber at the logging site and the working body of harvesters (processors) to carry out such harvesting will increase the share of logging waste at the logging site. The design of the new harvesting (processing) head does not require significant financial expenses for its upgrading. The design of the new harvester head allows the logs to be barked in strips along the entire length. Additional drive of working mechanisms is not required. This will enable the implementation of rough debarking with minimal financial costs already at the stage of logging operations and increase the efficiency of logging and timber processing enterprises.

ESTIMATING 3D FOREST STRUCTURE USING LIDAR DATA FOR RISK ASSESSMENT OF SLOPE COLLAPSE CONSIDERING ROOT SYSTEM DEVELOPMENT

Abstract. In order to predict slope failure, it is useful to be able to grasp the state of root system development of trees by remote sensing techniques. Since there is a positive correlation between tree diameter and root system development, a field survey to verify how predictable the number of trees and breast height diameter can be from LiDAR data was conducted in the Abukuma Mountains Granite Area and the Yamizo Mountains Jurassic Sedimentary Rock Area. It was carried out for the logging site. As a result, it was found that the relative tree density can be determined and that the laser transmittance and the sum of cross-sectional areas of the trees are inversely correlated.

Machine vision based waterlogged area detection for gravel road condition monitoring

ABSTRACT When assessing forest road conditions, information about waterlogged areas on gravel roads brings high practical value when used as an indicator for road wear. Around these perimeters, lowered binding forces of the construction material reduce the stability of the road, which induces accelerated road damage. When a road is actively used to access a logging site under humid weather or thawing conditions, road wear can build up fast and make further use of the road critical. In this study, a deep learning algorithm was trained to test the detection of a combined observation of waterlogged appearances on forest roads from video and image data, collected from a passing vehicle’s perspective. The training of a YOLO v5s model achieved an F1-score of 0.59 and shows the applicability of this approach with high confidence of detection. Evaluating further training characteristics such as precision, recall, and the object size-related detection confidence reveals challenges for a successful application in terms of undetected objects, variation of objects in the training step, the required amount of training data and the object distance focused.

Damage and Tolerability Thresholds for Remaining Trees after Timber Harvesting: A Case Study from Southwest Romania

The present study analyses the damage of remaining trees after timber harvesting from 24 logging sites from southwest Romania. The purpose was to establish tolerability thresholds within which damaged trees recover in a short amount of time, reducing the possibility of further rot apparition and tree health deterioration. Observations were resumed after the growing season had passed. Healed damage was analysed in regard to damage type, width, orientation and tree circumference. By using the ratio between the width of healed damage and the circumference of trees as experimental variants, equations were elaborated to determine the tolerance threshold of trees in logging. This is expressed as a maximum value between the damage width and the damaged tree circumference for which the damage is curable. The correlation between the circumference and the abovementioned relation was analysed, and differences between the values of the analysed relation for different cardinal orientations of the damage were statistically tested. The value of this ratio, which can be considered a tolerance threshold for trees in logging, records values of 0.09 (for thinnings, for cuttings to increase the light availability for regeneration and for final cuttings from shelterwood systems) and 0.10 (for first-intervention cuttings, as well as preparatory and seed cutting from shelterwood systems or selections systems).

The Design and Realization of Intelligent Classification of Rock Samples Based on Convolutional Neural Network

With substantial improvement of computer hardware performance and rapid development in artificial intelligence research, artificial intelligence recognition technology has been applied in many industries, such as medicine, education, industry, commerce, exploration and other professional fields. In geological exploration, the classification and identification of rock samples is an important link in geological analysis. However, the traditional manual classification faces high economic and time costs, and is vulnerable to subjective judgment which may affect the recognition result. In order to avoid these problems, research is carried out on the specific application of computer-based artificial intelligent recognition technology in recognition of rock images and the performance difference caused by different parameters. The sample data used in the experiment are lithic fragments and core samples capture by industrial cameras at the well logging site, with a total of 350 images which cover 7 types of rock samples. In this paper, the 350 rock images are divided into training set, validation set and test set based on certain proportion. Since the data set is too small and the quantity of each type of rocks varies, data augmentation technology is used to expand the original data before training. During the expansion, the quantity of different types of rock samples are balanced. At the training stage, the open-source deep learning framework Pytorch is used to construct a multi-layer convolutional neural network for learning and classification training on the training set. The best accuracy rate in the tests reaches 82.86%. However, the recognition rate is poor for stretched images, which required further optimization of the network structure. Based on the experimental results, the optimization orientations of neural network training are summarized, including improvement of data set quality, optimization of network structure, and adjustment of training strategies. In order to display the experimental data as intuitively as possible, data visualization tools such as PlotNeuralNet and Python matplotlib are used in the experiment for specific visualization work, after which the experimental data and model effect are analyzed from the perspective with combination of data and images.